New J. Phys. 26 (2024) 105002 [Link]

1088/1367-2630/ad42c6

PAPER

Tutorial on laser locking techniques and the manufacturing of

OPEN ACCESS
vapor cells for spectroscopy
RECEIVED
25 January 2024 Max Mäusezahl, Fabian Munkes and Robert Löw∗
REVISED
5. Physikalisches Institut and Center for Integrated Quantum Science and Technology, Universität Stuttgart, Pfaffenwaldring 57,
10 April 2024
Stuttgart, 70569, Germany
∗
ACCEPTED FOR PUBLICATION Author to whom any correspondence should be addressed.
24 April 2024
E-mail: [Link]@[Link]
PUBLISHED
4 November 2024
Keywords: atomic physics, molecular physics, laser locking, laser stabilization, Pound–Drever–Hall, FPGA, vapor cells

Original Content from

this work may be used
under the terms of the Abstract
Creative Commons
Attribution 4.0 licence. This tutorial provides a hands-on entry point about laser locking for atomic vapor research and
Any further distribution related research such as laser cooling. We furthermore introduce common materials and methods
of this work must
maintain attribution to
for the fabrication of vapor cells as a tool for this research. Its aim is not to be exhaustive, but rather
the author(s) and the title to provide an overview about the possible techniques that are actively employed in labs today.
of the work, journal
citation and DOI. Some critical parameters of locked laser system for use with thermal atomic vapors are introduced
and discussed. To exemplify this, we describe a versatile locking system that caters for many of the
needs we found during our research with thermal atomic vapors. We also emphasize the
compromises we took during our decision-making process.

List of acronyms

ALD atomic layer deposition

AOM acousto-optic modulator
ASG arbitrary signal generator
BS beam splitter
CTE coefficient of thermal expansion
DAVLL dichroic atomic vapor laser lock
DBR distributed Bragg resonator laser
DFB distributed-feedback diode laser
ECDL external-cavity diode laser, sometimes also called extended-cavity diode laser
EIT electromagnetically induced transparency
EOM electro-optic modulator
FM frequency modulation
FMS frequency modulation spectroscopy
FPGA field–programmable gate array
FSR free spectral range
FVC frequency to voltage converter
FWHM full width at half maximum
IQ in-phase and quadrature
LO local oscillator
MEMS microelectromechanical system
MTS modulation transfer spectroscopy
OFC optical frequency comb
OFLL optical frequency lock loop
OPLL optical phase lock loop
PBS polarizing beam splitter
PD photo diode

© 2024 The Author(s). Published by IOP Publishing Ltd on behalf of the Institute of Physics and Deutsche Physikalische Gesellschaft
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

PDH Pound–Drever–Hall
PFD phase frequency detector
PID proportional-integral-derivative
PM phase modulation
PSD power spectral density
RAM residual amplitude modulation
RF radio frequency
RP RedPitaya STEMlab 125-14
SAS saturated absorption spectroscopy
SHG second harmonic generation
SNR signal-to-noise ratio
TDLAS tunable diode-laser absorption spectroscopy
TiSa titanium-sapphire laser
ULE ultralow-expansion
VCSEL vertical-cavity surface-emitting laser
VECSEL vertical-external-cavity surface-emitting laser
WM wavelength modulation
WMS wavelength modulation spectroscopy

1. Introduction

Quantum optical experiments with thermal or cooled atomic vapors require controlled access to the
electronic level structure of the atomic system. Since the interaction between these electronic states emits or
absorbs electromagnetic radiation, this often means that precisely tuned lasers are employed to prepare and
probe atomic ensembles. The toy example of a two level, dipole allowed atomic transition between a ground
state |g⟩ and an excited state |e⟩ can be used to derive some principle requirements for such a radiation
source: It needs to match the frequency ν0 = |E|g⟩ − E|e⟩ |/h with a linewidth ideally smaller than the
linewidth ∆ν of the atomic transition. This ensures that the atoms effectively couple to the electromagnetic
field. Commercially available chip scale atomic clocks achieve a clock stability of ∆ν/ν0 which can be as small
as 5 × 10−11 [1], which is a tough requirement for the laser stability in the presence of external influences [2].
The specific case of thermal vapors sometimes relaxes the requirements regarding the precision of the
absolute laser frequency ν and its linewidth ∆ν in comparison to experiments with cold atoms. This is due
to the Doppler-shift and associated Doppler-broadening resulting from the thermal Maxwell–Boltzmann
velocity distribution. Within the Doppler-width ∆D , a significant number of atoms can be expected to fall in
a velocity class v that compensates any residual detuning ∆ = kl · v (laser wave vector kl ). Note however, that
this is only true as long as only a single transition is investigated. In multi level excitation schemes, a
sub-Doppler selection based on the interplay between the individual transitions is possible.
In either case the so-called free-running stability of available laser systems is usually not sufficient for
experiments with atomic vapors. They suffer from a non-constant center frequency ν movement called ‘drift’
due to varying environmental conditions or aging. Additionally, the arising laser linewidth ∆ν is broadened
due to internal noise processes or limited spectral selectivity of the laser cavity. Both issues can be solved by
‘laser locking’ which is a coined phrase for providing feedback to the laser in a feedback loop to stabilize its
frequency (sometime also called ‘laser stabilization’ or ‘frequency stabilization’). A reduction of laser
linewidth is usually referred to as ‘linewidth narrowing’.
Laser locking is a tool commonly employed by many laboratories working with lasers in the fields of
atomic physics, molecular physics, or optical metrology. Commercial turn-key solutions for many
applications are readily available. Nevertheless, the technology is still subject of current research on a pursuit
towards smaller packaging [3–7], even narrower linewidths [2, 8–11], better long-term stability [12] and
usability features like AI support [13], autonomous or automatic operation [5, 13–20] and cost-efficiency
[19, 21–28]. This tutorial neither aims to be an exhaustive review of all possible locking schemes with their
respective advantages, nor to roll out the full underlying theory of feedback loops in general or their specific
application with laser systems. It is much rather meant as a practical guide with a brief theoretical foundation
for the undergraduate or graduate student while being exposed to locked lasers for the first time. It is also
targeted towards senior researchers, who are planning experiments with hot atomic vapors and looking into
the manifold options that exist today—possibly based on an existing variety of laser systems and controllers,
which they wish to upgrade.
Introducing or getting introduced to laser locking in an accurate and adequately formal way without
getting lost in the many details is not any easy task and can feel overwhelming. It is not a narrow subject and
much rather sits at the interface between control theory, RF operation, and optics. Each of those subjects is
worth dozens of textbooks on its own. Luckily, a wide range of well-documented solutions exist, and it is
certainly possible to set up intricate experiments without dedicating years of learning.

2
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Figure 1. (a) Basic setup to perform an Allan frequency deviation measurement between two lasers. The lasers have a radio
frequency (RF) frequency offset that is observed as an electric RF beat note and can directly be sampled and analyzed using e.g. an
oscilloscope or spectrum analyzer. If both lasers have the same frequency, an additional frequency can be introduced using a
modulator (see section 4). (b) The textbook Allan deviation plot shows the Allan deviation σ(τ ) over measurement time τ in a
log-log plot. The slopes in the log-log representation as indicated above the six highlighted sections hint towards the origin of the
noise dominating the fluctuation on the respective timescale (from left to right: quantization noise, white noise, correlated noise,
bias instability, random walk rate, overall trend). An actual measurement would not show the different sections in this clarity.

We purposely suggest a larger number of references, which could serve as a starting point for additional
reading on the individual methods in the various sections. Moreover, there are a number of existing review
articles and books covering the topic of laser locking. Some of them are technically mostly outdated due to
their age but still feature a solid introduction into the fundamental principles [29–35]. Recent comprehensive
publications on the matter include the articles by Hall et al [36], by Taubman [37], and by Wu et al [38] as
well as the respective chapter in a book by Fang et al [39]. Additionally, there are in-depth publications
covering a specific set or aspect of laser locking techniques [40, 41].

2. Characterizing laser frequency fluctuations

Previously, we introduced the ad-hoc terms of external noise, laser linewidth, and center frequency drift. As
physicists, we might already have an intuitive sense of the meaning and delicate interplay between these
quantities, e.g. how stronger acoustic noise could lead to a larger laser linewidth. A rigorous treatment and
precise definition of these details is possible by observing frequency domain spectra or time domain
variances [42, 43]. There are numerous standardized quantities which can be explored [44], of which we will
introduce the widely used Allan variance in time [45], and PSD in frequency [35].
We start from the oscillating, instantaneous optical field amplitude E(t) (or other source amplitude),
which is typically not directly accessible in an optics
´ ∞ experiment. This field has an instantaneous frequency
f (t) and a total, average power P = α limT→∞ −∞ |E(t)| dt (with some factor α depending on the type of
2

the field).
The PSD PE (f) of this field E is then defined using the Fourier transform of the autocorrelation according
to the Wiener–Khinchin theorem, such that the total power within a certain frequency interval f1 ⩽ f ⩽ f2 is
given by the integral over this frequency range (again with some normalization factor β):
ˆ f2
P[ f1 ,f2 ] = β PE (f ) df. (1)
f1

In practice, it is this integral formula that makes the PSD useful to understand short-term effects of fractional
power located close to the desired optical target frequency. Since the PSD of lasers used in atomic physics is
usually narrowband, a FWHM linewidth of this PSD is often given in literature as a single figure of
performance. The aforementioned drift of the laser frequency describes a non-constant peak frequency of the
PSD when repeating the measurement over time. The PSD can be directly optically measured using a
scanning analysis resonator, an autocorrelator, or a beat-note analysis incorporating a second reference laser
[35].
The Allan variance on the other hand is understood as a measure of the long-term performance of an
oscillator, i.e. the long-term stability of the laser frequency. Like the standard variance it measures a statistical
spread of a set of numbers (in this case frequency differences or phase differences) but does not diverge for
common noise sources such as flicker noise. A concrete measurement captures N frequency deviations ∆f n
with n ∈ {1, 2, . . . , N} between two laser systems from a beat-note signal, see figure 1(a), averaged over a
short observation time τ 0 at certain times tn = nτ0 . We assume zero dead time between measurements here.
These are often converted to dimensionless fractional frequencies yn = ∆fn /f0 − 1. The Allan variance σy2 (τ )

3
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

is then calculated as

N−1
∑
1 2
σy2 (τ ) = (yi+1 − yi ) . (2)
2 (N − 1)
i =0

This can be intuitively understood like a standard variance—notice the general form of a time average of
squared deviations—that however takes into consideration deviations between successive samples instead of
deviations from the overall mean. Conventionally one plots the square root of the Allan variance, called Allan
deviation σy (τ ), on a log–log plot over the total measurement time τ . Such plots are exceptionally rich to the
experienced eye [42], so that we only give two brief starting points for interpretation here: a small value of
σy (τ ) at a certain time τ indicates a small root-mean-square frequency deviation between the two lasers over
this time span. The slopes appearing in such a plot are an indicator for the underlying type of noise process,
which follow specific power laws as shown in figure 1(b). Today, use of the original Allan variance has been
largely superseded by extended and improved variants like the overlapping Allan variance or the modified
Allan variance [44].
We describe how both the PSD and the Allan deviation are measured in practice in section 6.4.

3. The laser feedback loop

It is impossible to talk about laser locking without a basic understanding of feedback loops and control
theory. Control loops are hidden in many of our modern devices, e.g. in each switched-mode USB power
supply in order to deliver a precise voltage to mobile devices. They are even more common to experimental
physicists, who are dealing with a plethora of heterogeneous systems, which need to be held within tight
operational windows, such as a vapor cell that is kept at a specific temperature by a thermal controller. This
section introduces their most important concepts in the context of laser locking specific issues. A broader
and more detailed introduction into ‘feedback for physicists’ can be found in the identically named [46].
The first step to understand, how the frequency of a laser can be stabilized, is to recognize, why this
becomes necessary at all. A textbook laser consists of an optical resonator (called cavity here), an active
medium, and an energy pump, which define the emission spectrum. In reality, there are environmental
influences that result in a non-constant laser frequency over time owing to temperature, humidity and air
pressure fluctuations, mechanical vibrations, aging effects, and even external electromagnetic fields. These
lead to varying refractive indices, changing lengths, and emission properties making the overall system highly
unpredictable. Therefore, a ‘passive stabilization’ of the laser system is mandatory even when active laser
locking is employed. As [47] puts it: ‘Generally, the first step to improve the laser performance is to make the
cavity construction insensitive to these external effects.’ It must be stressed that it might be impossible to
reliably lock a laser, if the overall mechanical structure is too unstable.
There is however a point, at which it becomes technically infeasible to add more passive stability, as this
usually comes with a larger build volume and an excessively increasing cost. In such cases, the initially
‘open-loop’ system is expanded by a feedback loop, which monitors the actual, instantaneous laser frequency
ν(t) through an optical frequency discriminator (figure 2(a)). The stability issue is then transferred to a
comparison between ν(t) and a known stable frequency reference νset . Many optical frequency
discrimination devices have been developed, some of which are covered in section 5.
In each case the optical frequency discriminator produces an error signal e(ν), which carries the
information about momentary state of ν(t) compared to νset . This signal must thus contain both the sign
(i.e. ν < νset or ν > νset ) and the magnitude |ν − νset | of any deviation. It is therefore often encoded as a
voltage signal roughly proportional to ν − νset . This ideal case is rarely achieved, but a decent approximation
is sufficient for a suitable controller (also called filter due to its inner working principle). By setting a number
of control signals s(t) of the laser, the controller counteracts the deviation as reported by the optical
frequency discriminator. Figure 2(a) also includes various sources of noise and disturbances which are
imperfections of the system. It is ultimately the goal of the operator to tune the various filters and signal
amplitudes such that external noise sources are effectively suppressed, without introducing additional noise.
Reasoning about the properties of the overall system in the time domain as a control theory problem is
usually impractical and thus much rather performed in the frequency domain f of all noise sources ξ(f) [36].
In this representation (figure 2(b)), controller and laser form a frequency dependent system G(f) which
responds to feedback H(f ) in the presence of certain noise ξ(f). Notice, that G and H generally are complex,
normalized transfer functions which include phase-lag of the various components, and that the frequency f
is not an optical frequency close to ν, but a much smaller, unwanted oscillation frequency component

4
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Figure 2. Two schematic representations of a laser locking feedback loop, both inspired by [36]. (a) A model showing the key
components of the system, starting with a laser that emits a certain noise-disturbed and time-dependent frequency ν(t), of which
a portion is split off and sent to an optical frequency discriminator. This compares it with some setpoint frequency νset and
produces an error signal e(ν) that is processed by a controller to counteract any disturbances via control inputs s(t) of the laser.
(b) A schematic view of the same system from a control theory standpoint, where the action of each component is represented by
a complex valued, frequency f domain transfer function. H(f ) describes the amplitude response and phase-lag of the feedback
sensor system, while G(f ) represents both the controller and the laser.

disturbing the operation. The new overall system has a transfer function [36]

G (f )
L (f ) = , (3)
1 + G (f ) H (f )

which can be tuned through the controller settings and the feedback sensors. At a first glance, making the
amplitude of H as large as possible would in theory eliminate all noise, however a closer look identifies time
lag (and therefore phase lag) of the feedback H(f ) as the ultimate limit. For |GH| = −1 self-reinforcing
positive feedback (out of phase by 180◦ , also called feed-forward) is achieved. In these circumstances the
laser emission frequency would start to oscillate (modulate) at such frequencies on its own, which is the
contrary of a stabilized laser. Therefore, feedback loops are carefully designed and then hand-tuned to
strongly counteract unwanted frequency components, e.g. acoustical frequencies in the kHz range with high
gain, while the inevitable frequency, at which the phase lag becomes larger than 180◦ is appropriately
dampened. It follows, that each feedback loop has a usable feedback bandwidth fmax above which frequency
noise cannot be sufficiently reduced. As a rule of thumb (assuming a reasonable choice of the control
parameters) a high bandwidth is necessary to reduce high-frequency noise and narrow the laser linewidth.
Small feedback bandwidths could nevertheless be sufficient to eliminate drifts.
A basic setup might use a PID controller in combination with various low-pass filters to suppress these
resonance points: A proportional controller that directly counteracts e(ν), an integral part that corrects any
residual error ⟨e(ν)⟩t (‘proportional droop’ at low frequencies [46]), and a derivative part that operates at
high frequencies, trying to ‘predict’ rapid changes of the system e.g. due to a singular external disturbance.
Modern commercial controllers with high bandwidth offer a wide range of settings such that ‘optimal
control’ (a coined term for the ‘best’ possible settings) might only be possible by employing computer
simulations and rigorous measurements. FPGA based systems like the one we use in our lab are seeing
increased usage in commercial and self-made solutions over conventional analog systems. This is due to their
exceptional versatility and cost-effective integration of fast mathematical operations like IQ demodulation
with high bandwidth digital filtering into a single system [48–50]. Controllers can also include usability
features like automatic tuning [51] or automatic locking if the frequency is lost (re-locking). A common
manual procedure to lock a so far unknown laser, called ‘tuning’ the controller, is presented in section 6.2.
Beyond the (electrical) controller itself, the response of the laser to its control parameters s including all
delays is described within G(f ) in (3). Consequently, a high-bandwidth controller requires a high-bandwidth
response of the laser itself regarding the control signals s(t). Laser spectroscopy with atoms and molecules
sees common use of various laser types that have different characteristics in this regard:

• ECDLs are commercially available with free-running linewidths of few 100 kHz and tuning over tens of nm
[52]. High bandwidth control (many MHz) is possible through the pump current which has the unwanted
side effect of output power modulation. Therefore, it is usually combined with the mechanically limited,
lower bandwidth piezoelectric length tuning of the external cavity (few kHz). Low bandwidth control based
on diode temperature is mostly seen as a part of the passive stability of the system, i.e. it is not part of the
laser frequency feedback loop and rather controlled by a different controller using a dedicated temperature
sensor [53, 54].

5
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

• Distributed-feedback diode lasers (DFBs) and the closely related DBRs are similar but mechanically simpli-
fied and thus cost reduced alternatives to ECDLs with less tuning range (few nm) and larger free-running
linewidths, that can nevertheless be greatly reduced through laser locking and the accompanying linewidth
narrowing [55]. Again, high bandwidth control (many MHz) is available through the pump current while
large wavelength changes and low-frequency control (few Hz) is provided through the chip temperature.
• Fiber based single-frequency lasers are typically either a DFB or a DBR implemented inside a fiber and thus
feature the same control mechanisms [56].
• Dye lasers have historically been the first widely tuneable, high-power lasers systems available at a wide range
of wavelengths that could be simultaneously narrowed to linewidths below 1 MHz [57, 58]. Coarse tuning is
performed using cavity length tuning, which can be part of the feedback loop. Finer mode-hop free tunab-
ility is then typically achieved by placing a dispersive, piezoelectric actuator driven (several 10 kHz feedback
bandwidth) wavelength selector such as a Littrow configuration grating, a Fabry–Pérot interferometer, or
a birefringent filter such as a Lyot filter inside the laser resonator [59]. Higher feedback bandwidth can be
achieved using acousto-optic modulators (AOMs) or electro-optic modulators (EOMs) inside the cavity
[60].
• Solid state Ti:sas are currently the predominant femtosecond laser source due to their large spectral band-
width and tuning range. They are built similar to dye lasers yet overcome many of their detrimental prop-
erties and can be tuned by intra-cavity elements just as described above [61, 62].
• VECSELs have recently seen increased use in the quantum optics community as they bridge a gap between
the low-power but easily tuned semiconductor lasers and the high-power solid state lasers [63]. While elec-
trically pumped VECSELs exist, optically pumped VECSELs currently have a technological edge. Thus, the
tuning mechanisms in available VECSELs systems are comparable to dye systems and Ti:sas using cavity
tuning and intra-cavity elements [64].

Finally, it is possible to add an arbitrary, posterior high-bandwidth control to any laser by adding an optical
frequency modulator to the output, which can then serve as a control point in the feedback loop [32].

4. RF laser modulation techniques

We write the temporally changing electrical field vector E(t) of a linearly polarized, monochromatic plane
wave at a fixed position as
[( ˆ t )]
E (t) = e (t) E0 (t) exp i 2π ν (τ ) dτ + φ (t) (4)
0
= e (t) E0 (t) exp [i (2π νt + φ (t))] . (5)

Modulation refers to the periodic variation of either the polarization unit vector e(t), the amplitude E0 (t),
the frequency ν(t), or the phase φ(t) with a certain modulation frequency fmod . Notice that (5), which is
commonly found in textbooks, is only true in case of constant frequency ν. Usage of optical RF modulation
within the context of laser locking falls broadly into one of three applications:

(i) frequency sideband generation as required by many frequency discrimination techniques,

(ii) offsetting, postlaser-correcting [65], or sweeping an already locked laser frequency, or
(iii) adding an arbitrary, posterior, high-bandwidth control within the feedback loop, when it is not easily
available otherwise [32, 66].

AOMs directly shift the frequency of the diffracted light by some multiple of the applied acoustic wave
frequency and can thus be used as frequency modulators or frequency shifters with sub GHz bandwidth.
EOMs exist as phase, polarization, and amplitude modulators, of which phase modulators with multi GHz
bandwidth are particularly common to laser locking.
Notice that, for most practical sideband generation purposes, FM and PM can be used interchangeably,
as they are specific cases of angle modulation. Therefore, an additional modulator might be omitted in favor
of direct laser current modulation which leads to FM. The latter comes at the price of frequency sidebands
present in the main beam, which is sometimes acceptable, e.g. if the main beam is subsequently filtered
through a cavity as part of a SHG process, or if it solely serves as a reference for other lasers. Nevertheless,
AOMs or EOMs will usually be found in separate side beams dedicated to laser locking, while the main beam
stays free of sideband modulation and is controlled by means of any other aforementioned control signal.
Sinusoidal FM or PM in (4) are given by

ν (t) = ν0 + ν̂mod cos (2π fmod t) , with 0 ⩽ ν̂mod ⩽ ν0 and, (6)

6
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Figure 3. (a) The temporal shape of an unmodulated, normalized, sinusoidal electric field amplitude in comparison to an FM
sinusoidal signal with the same carrier frequency ν and fmod = ν/10. (b) The relative amplitude of all sidebands of an
unmodulated, sinusoidal signal and an FM signal. (c) The relative amplitude per sideband over different modulation indices for
the carrier and the lowest three sidebands. (d) The spectral power of an unmodulated, sinusoidal signal and an FM signal
normalized to the total power. (e) The spectral power per sideband over different modulation indices for the carrier and the
lowest three sidebands. Throughout the plotted region the combined power in all other sidebands q > 3 is less than 4%.

φ (t) = φ0 + φ̂mod sin (2π fmod t) (7)

respectively. In both cases the modulation leads to the creation of sidebands at multiples of fmod , i.e. at
ν0 ± q fmod , q ∈ {0, 1, 2, . . .}. It is a common mistake to assume that the power in all higher order sidebands
q > 1 is negligible. While EOMs might inherently limit the achievable modulation index, this is typically not
the case for laser current modulation. Yet, even for a perfectly sinusoidal modulation, all those sidebands are
present with some nonzero amplitude (see figure 3).
It is possible to directly observe the optical power in the modulated sidebands using an optical resonator
with a linewidth smaller than the modulation frequency. Modulating the length of the resonator will lead to a
transmission signal like figure 3(d) on an oscilloscope. This however hides the relative phase between the
sidebands, which is schematically shown in figure 3(b). All negative odd order sidebands have a phase shift of
π with respect to the carrier represented as a negative sign in the amplitude. All other (especially the negative
even order) sidebands stay in phase with the carrier. This phase difference between the lowest order
sidebands q = −1 and q = 1 is the foundation of many phase sensitive detection mechanisms discussed
below. Notice that overdriving the modulation strength with the hope of improving signal-to-noise
performance in such cases, will not contribute to the signal strength.
It turns out that the spectral power which is transferred into each sideband is conveniently described by a
modulation index m which depends on the peak frequency or phase variation and the modulation frequency,

mFM = ν̂mod /fmod (dimensionless), (8)

mPM = φ̂mod (radians). (9)

Note, that while we follow the convention of [67], the term modulation index is sometimes used
interchangeably with the term modulation depth [68, 69]. The precise amplitude of each sideband of order q
is given by Bessel functions of first kind as Jq (mFM ) or Jq (mPM /rad), plotted in figure 3(c) [67]. Squaring
these amplitudes results in the sideband spectral power as plotted in figure 3(e). As there are always two
symmetrical sidebands except for the carrier at q = 0, the total power is conserved due to
∑∞
J20 (x) + 2 q=1 J2q (x) = 1 (this also means that PM does not add power to the field) [70]. Finally, since any
arbitrary periodic signal can be understood as a linear Fourier composition of multiple sinusoidal terms, the
presented relations can be transferred accordingly.

7
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Table 1. Overview over most frequency discrimination methods discussed in the individual subsections in the main text based on the
information given in the references of these descriptions. The ‘reference’ column indicates which component or oscillator serves as a
frequency reference. Atomic references are absolute, while cavities and interferometers need to be calibrated with an external standard.
The ‘RF’ column lists, whether RF modulation and demodulation of laser light is required. Typical bandwidths have been extracted from
the various references and can only give a rough indication, while specific modifications might allow or require much higher or lower
values. The last column shows, whether the technique is of interferometric nature, i.e. whether it is based on the coherent interaction
between multiple fields on the detector.

Method Reference RF Typical bandwidths (kHz) Interferometric

Side of fringe Various no >10 no

Dither lock Various no >100 no

Pound–Drever–Hall (PDH) Cavity yes >1000 yes

Hänsch–Couillaud Cavity no >100 yes

Saturated absorption spectroscopy (SAS) Atomic no >10 no

Polarization spectroscopy Atomic no >100 no
Dichroic atomic vapor laser lock (DAVLL) Atomic no >100 no
FM spectroscopy (FMS) Atomic yes >1000 yes
Modulation transfer spectroscopy (MTS) Atomic yes >100 yes

(Scanning) transfer cavity Laser varying

Optical frequency lock loop (OFLL) Laser yes >100 yes
Optical phase lock loop (OPLL) Laser yes >10 yes
Optical frequency comb (OFC) Laser yes >1000 yes

Interferometer/Wavemeter Various no <1 yes

5. Overview of common frequency discrimination techniques

While the selection of a suitable electronic controller as discussed in the previous sections is undoubtedly
important for the eventual success of the laser lock, there are a number of commercial and open-source
solutions that work interchangeably for a wide range of applications. This section now describes the optical
part of the laser lock, which is typically a deliberate choice based on a case-by-case evaluation. In this tutorial
we try to provide an overview and classification of various commonly deployed optical frequency
discrimination techniques. This selection as outlined in table 1 is based on appearance in recent publications
but also includes some less common variations mainly for their historic significance or educational value in
the overall context.
The central goal of each optical frequency discriminator is the creation of a typically dispersive error
signal e(ν) with respect to a target laser frequency νset that can be fed into an electronic controller. This is
done by comparing the actual laser frequency with a known absolute or relative frequency reference.
Absolute references include atomic and molecular transitions which lend themselves particularly well as a
source of stability when working with their vapors and form the foundation of modern time standards [71].
Optical cavities, interferometers, and reference lasers are relative references, that will drift over time if
exposed to environmental changes, but can be stabilized and calibrated to serve as a fixed, almost absolute
frequency if necessary.
Another classification of each technique can be made based on the frequency range [νmin , νmax ], where the
sign of the error signal properly matches the actual sign of ν with respect to νset . This so-called capture range
is usually limited due to bandwidth caps or periodic signal features. Additionally, each technique has a
maximum usable bandwidth, either due to inherent limitations such as a modulation frequency, due to
sensor bandwidth, or due to a finite sampling rate. For many techniques RF modulation and demodulation
of the laser light is required. While this often leads to a reduction of noise at the modulation sidebands
comparable to a lock-in amplifier, it also comes at the expense of additional AOMs, EOMs, and RF hardware
and usually limits the bandwidth to the chosen modulation frequency. Bandwidth alone is yet not sufficient,
as there must be strong frequency discrimination (expressed as a slope of e(ν) around the target frequency).
If the spectral reference feature is broad, this leads to a weak response for small frequency changes compared
to the overall electric noise in the system and therefore a bad SNR. Finally, it is possible to classify each
scheme based on the presence of interferometric coherence for signal generation: Some solely depend on a
single, incoherent intensity measurement, while others interfere multiple fields coherently [72].

5.1. Locking to an optical cavity

An optical cavity or resonator like shown in figure 4(a) is formed by two (or more) mirrors, between which a
resonant electromagnetic field can build up. In this example a quarter wave plate is added in front of the

8
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Figure 4. A linear optical cavity (drawing and plot inspired by [37, 73]). (a) Schematic drawing of the cavity with length L
including photodetectors for the transmitted intensity It and the reflected intensity Ir . Assumes a linearly polarized input beam
such that the ideal cavity and a quarter wave plate aligned under 45◦ to the polarization axis will lead to the reflected beam being
also reflected at the polarizing beam splitter (PBS). (b) From top to bottom: The transmitted intensity, the reflected intensity
(both normalized to It + Ir ), and the phase shift of the reflected field ∆φr all at a finesse F = 20.

cavity such that the incident linearly polarized beam is converted to a circular polarization and then, after
reflection, again to a linearly polarized beam with a 90◦ rotated polarization plane. This clearly separates the
reflected beam from the incident beam. We assume no energy loss due to absorption in the following
discussion for simplicity, such that the transmitted intensity It and the reflected intensity Ir sum up to the
incident intensity Ii = It + Ir . Normally, the light will be almost completely reflected at the first mirror, and
only a fraction finds its way into the cavity (off resonance Ir ≈ Ii ). If however the resonance condition
Mc = 2νL is fulfilled for any positive integer M (longitudinal mode number), the circulating light field builds
up inside the cavity over successive cycles. Eventually it completely destructively interferes with the reflected
light field and the transmitted intensity equals the incident intensity (on resonance It ≈ Ii ).
The frequency difference between two neighboring resonances at M and M + 1 is called FSR and given by
δνFSR = c/(2L). The exact shape of the transmitted and reflected amplitudes and phases can be derived as a
steady-state problem and is included in many standard textbooks [73]. For the purposes of this tutorial it is
sufficient to discuss the prominent features as plotted in figure 4(b) qualitatively: The transmitted and
reflected intensity signals are periodic and produce an identical Lorentzian line profile for each longitudinal
mode in distances δνFSR . Each peak has a certain FWHM δνFWHM which is approximately (for R > 0.5)
related to the intensity reflectivity R and length L via the finesse F like
√
δνFSR π R
F= ≈ . (10)
δνFWHM 1−R
Consequently, a smaller FWHM is usually attributed to a high reflectivity R, when the length L is governed by
the need for a specific FSR. The phase shift ∆φr of the reflected light field compared to the incident light
field shows a strong dispersive line shape around the resonances. The finesse F can be understood as a
quality factor of the optical resonance and is proportional to the mean number of reflections before a photon
leaves the cavity. This is related to the fact, that the cavity field has a certain build-up time before the
transmission reaches the plotted steady state. Often high finesses and therefore long build-up times are
required to achieve extremely narrow features. As this would undermine the idea of a fast, high bandwidth
feedback response most laser locking systems are build using the reflected field [37]. Other shapes of cavities
with more than two mirrors including the circulating bow-tie setup are used in some situations: The bow-tie
configuration for example has a clear geometric separation between the incident and reflected light field.
However, the presented fundamental theory does not change with the geometry.
An optical cavity made from spherical mirrors also acts as a filter regarding the transversal mode profile
of the incident laser field. The overall coupling which manifests itself in the peak intensity transmission or
the contrast of the reflected signal does therefore also depend on a transversal mode matching, e.g. to a
Gaussian mode profile. For a high overall coupling to a cavity the cavity must match the longitudinal modes,
longitudinal impedance, and transversal modes [74]. These must be considered during the design phase as
certain geometries and mirror curvatures might demand impracticably large or convergent beam shapes.
The open-source software REZONATOR2 [75] can be used for Gaussian beam type cavity calculations, while
the other properties can sometimes be only approximated by the manufacturers, especially if highly reflective
thin-film coatings with R ≫ 0.99 are applied.
In summary, the advantage of cavities as a reference for laser locking lie in their simplicity, the principle
availability at all wavelengths, and their periodicity. This means that a laser can be locked essentially at any
desired frequency (maybe with the help of offset or sideband locking as discussed below). They can also have

9
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

extremely high finesses of 106 and beyond which makes them specifically suitable for linewidth narrowing
[76]. The drawback to all these methods is, that they are still based on a relative reference that can drift over
time and needs to be calibrated against an absolute reference. By using ULE spacer materials with low CTEs
inside a mK-stabilized, vibration isolated vacuum environment it is possible to reach a limit of material aging
and thermal noise of the mirror substrates [9, 77, 78]. Possible solutions include the use of different materials
with low thermal noise at cryogenic temperatures [2]. An up-to-date introduction to such ultra stable
cavities with practical insights into the laser locking aspect can be found in [79]. For practical applications it
can also be sufficient to maximize some observed signal without ever fully calibrating the absolute frequency
of the cavity modes.

5.1.1. Side of fringe lock

A typical first idea is to ‘simply’ lock the laser to the peak of an atomic transmission or a cavity resonance
with a setup as shown in figure 4(a). While the reflected signal in figure 4(b) shows a clear minimum to the
human eye, such a signal does not contain an information about the sign of a frequency deviation, i.e. the
signal intensity just increases in either direction making it impossible to precisely counteract. It is however
possible to use such a signal when locking to the side of the resonance, also called fringe [80, 81]. The
reference is given by a stable voltage Uset which defines a specific frequency compared to the photodetector
voltage Ur (ν) in reflection [34]. This has the immediate advantages of great simplicity and even versatility, as
a lock point can be chosen at an arbitrary frequency along the flank.
Side of fringe locking is still widely used in settings that require only a low-bandwidth stabilization of the
central laser frequency such as TDLAS [82], coherent communication, gas monitoring, and interferometric
ranging. When the laser is continuously scanned over a certain frequency range, the side of fringe idea can be
expanded towards a dynamically updated frequency-map or look-up table, which associates all data points
with certain frequencies and acts accordingly.
Nevertheless, such laser locks are rarely seen in recent quantum optics publications, since it is the only
presented technique in this tutorial, which directly transfers amplitude noise of the laser into frequency noise
in first order. The capture range is limited to half the flank of the chosen resonance and a narrower, more
frequency sensitive FWHM increases this issue even more. Additionally, it is often simply the goal to be on
top of the fringe—especially with atomic resonances—rendering a side of fringe lock largely impractical.
Creating a top of fringe lock error signal can be seen as the underlying challenge of all other presented
locking schemes.

5.1.2. Dither lock

Dithering as a general concept just means that some parameter—in our case the laser frequency—is
modulated with some low frequency fdither . Note that the name ‘dither locking’ has no universally agreed
usage throughout academic literature and might refer to a wide range of frequency and PM locking
techniques, sometimes limited to the ones involving a direct modulation of laser current. For this tutorial by
dither locking we mean a rather low fdither which is on the order of the spectral width at high modulation
index, such as described in [34, 37] and called WM in [83]. This is typically achieved via direct laser current
modulation but can also be done using AOMs. In this configuration the transmission signal of a (low finesse)
cavity can be used as an input, meaning that the setup looks like the PDH setup at much lower frequency
(figure 5(a)). A lock-in amplifier is then used to detect the change of phase between the modulation signal
and the observed oscillation, which then serves as the error signal for the controller.
The advantage over PDH lies in its lower frequency operation that can be achieved with much simpler
lock-in amplifiers. The low modulation frequency simultaneously sets a bandwidth limit, which might even
be worsened by the low-pass filtering of a cavity in transmission. Historically it was chosen in particular since
it can be easily modulated through the current input of any laser, which also leads to an undesired high
modulation index of the laser frequency throughout the rest of the experiment [8, 84–88].

5.1.3. PDH type lock

The PDH scheme was originally demonstrated with masers by Pound in 1946 [89] and later applied to lasers
in 1983 by Drever et al [90]. It appears to be an extension of the dither lock to higher RF modulation at much
lower modulation index at a first glance, but works due to a fundamentally different effect [69].
A PDH setup as shown in figure 5(a) consists of a laser, which is phase modulated at fmod , reflected by an
optical cavity, and observed on a photodetector with a higher bandwidth than fmod . The idea is to recover the
phase information of the reflected first order sidebands (figure 3(b)), which is only possible by an
interferometric technique at optical frequencies. The modulation adds sidebands at ν ± fmod that also get
reflected at the cavity. Since they lie away from the resonance, they are virtually completely reflected with no
significant phase shift and serve as the phase reference for the heterodyne interference signal. The carrier

10
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Figure 5. (a) The standard configuration for a PDH type laser lock, where the phase modulated and reflected beam is detected and
mixed (multiplier symbol) with the phase-shifted LO (indicated by the sine-wave and ∆ϕ symbol). After low-pass filtering the
error signal is obtained with is fed into the controller. (b) The Hänsch–Couillaud type lock is based on a polarization birefringence
of a cavity induced by some internal polarizer. This is observed by means of the difference signal between two photodetectors.

frequency ν gets only partially reflected with a strong phase shift imprinted for small deviations from the
resonance frequency (figure 4(b)). It is this phase difference in interference with the virtually unchanged
sidebands that leads to an amplitude modulation on the photodetector at fmod , which can be recovered
electronically by mixing (i.e. multiplying) with a reference fmod signal. After a low-pass filter (< fmod ) a
dispersive error signal is created and subsequently processed by the following controller.
One immediate advantage of the PDH scheme is its inherent balance or ‘nulling’ around the cavity
resonance, as both sidebands ν ± fmod contribute equally to the signal. This ideally results in a signal with no
offset, which is also not susceptible to intensity fluctuations of the laser. Furthermore, the detection at fmod
occurs in a regime much higher than common low-frequency technical noise sources [37]. Only a small
modulation index is necessary for a strong interference as only a small portion of the carrier gets reflected
[69]. The capture range of the method is set by fmod in either direction from the cavity resonance, at which
point the sidebands themselves will be on resonance and introduce another zero crossing.
The primary disadvantage of the PDH scheme lies in its mandatory RF electronics and modulation. PDH
detection additionally suffers from any RAM, that will produce an offset of the resulting error signal. Great
lengths are taken to minimize RAM, which is commonly introduced by EOMs, or make it at least a stable
contribution over time [39, 91]. Nevertheless, PDH is probably the most widely adopted technique presented
in this tutorial, both historically and today.

5.1.4. Hänsch–Couillaud type lock

Slightly before Drever and Hall published the PDH scheme, Hänsch and Couillaud introduced another,
polarization based frequency discrimination technique in 1980 [92]. Interestingly, their work is also
motivated by the 1946 publication of Pound [89] while being aware of the work by Drever and Hall from
private communication [92].
It is therefore not surprising that, similar to the PDH scheme, the Hänsch–Couillaud setup consists of a
cavity at which the laser beam is reflected while the reflected signal is observed (figure 5(b)). In difference
however, the cavity must contain a linear polarizer or some optical component which introduces different
properties depending on polarization such as a Brewster-cut or birefringent crystal. This effectively makes
the cavity as whole birefringent, leading to a phase shift of the reflected light orthogonal to the axis of the
internal polarization optic. On resonance, both components stay in phase, leading to an effective linear
polarization of the reflected light, while off-resonant light will lead to an elliptic polarization. Importantly,
the handedness of the ellipticity depends on the sign of the detuning between laser and cavity resonance [92].
Since this ellipticity needs to be detected, the reflected signal cannot be separated through the quarter wave
plate and PBS combination. Figure 5(b) proposes a separation using a non polarizing BS (thus losing at least
75% intensity), while recent applications are mostly found using ring cavities that exhibit a geometric
splitting between incident and reflected beam [93–95]. The detection is performed as balanced detection
between two photodetector voltages behind a quarter wave plate and PBS combination. This difference does
not depend on the azimuth angle of the ellipticity, but only on its magnitude and handedness [92]. It can
therefore be directly used as the error signal for the controller.
The major advantage of the Hänsch–Couillaud type lock over PDH is the absence of any RF components
and sideband modulation. Even though any sidebands would be strongly suppressed in transmission, this
makes it the method of choice in applications where no sidebands can be present at all, e.g. during SHG in

11
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

certain experiments [95]. Due to the balanced operation between two photodetectors, common mode
intensity noise is suppressed.
These two photodetectors on the other hand must stay precisely matched to prevent offsets of the error
signal [34]. The ‘nulling’ suffers from bad transversal cavity coupling. Every component inside the cavity
increases losses and reduces its finesse [37]. Compared to PDH, the detection occurs at a frequency of zero,
which typically features the largest inherent technical noise. Most importantly however, it lacks the wide
capture range of PDH, rendering locks to narrow cavity lines particularly challenging.

5.2. Locking to an atomic or molecular reference

While the previous section on cavities inevitably required the discussion of the stability and stabilization of
such devices themselves, atomic or molecular transitions are seen as the ultimate source of frequency stability
and absolute precision. Locking a laser to an atomic resonance instead of a cavity works not fundamentally
different either. Much rather, many of the previously stated ideas and schemes can be directly applied (and
get different names nevertheless). This however comes at the cost of versatility: Resonances can only be
found at specific frequencies and come in fixed strengths and linewidths. Moreover, atomic resonances are
not as stable as they seem to be at the first glance, being influenced by external electric and magnetic fields,
other atoms in their vicinity, as well as their own movement [37]. Consequently, laser locking techniques
using atomic standards can be roughly divided into two categories: First, there are compact vapor cell based
devices used to produce light which is to be interfaced with another sample of the same atomic species. In
this configuration the described linewidth and versatility limits are actually advantageous, since they are
intrinsically suitable for the application. The second category includes typically room-filling experiments
which opt to set frequency standards based on optical transitions in state-of-the-art optical atomic clocks
[96]. For the purposes of this tutorial, we primarily have the first category in mind, even though the
fundamental techniques are not strictly different but rather more elaborate in the second case. All schemes
discussed in the following section broadly fall under the umbrella of laser spectroscopy techniques which are
detailed in textbook literature such as the ones by Demtröder [97, 98].

5.2.1. SAS dither lock

Of the techniques presented in section 5.1, both side-of-fringe and dither locking can be directly applied to an
atomic resonance observed in transmission through a reference vapor cell [99]. This configuration however
suffers from inherent Doppler broadening, especially if additional heating beyond room-temperature is
required to achieve sufficient vapor densities. SAS refers to a setup, where the incident laser beam is split into
a probe and a pump beam, which then cross the vapor cell in a counter-propagating fashion [100–102].
Figure 6 shows two methods derived from SAS with additional polarization optics, whereas the most basic
form requires no additional polarizers and just one photodetector. Doing so creates so-called sub-Doppler
Lamb dips in the transmission spectrum at the frequency of the atomic resonances. Their specific shape and
linewidth is subject to beam-diameters, temperatures, angles, and external fields [101, 103].
Whilst SAS is most commonly just understood as a spectroscopy method, it is occasionally referred to as a
locking technique on its own in scientific publications. If done so, this mostly refers to a dither lock by means
of laser current modulation applied around a Lamb dip achieved through SAS [104–106]. Staying in line
with the naming conventions, the name WMS is also used to differentiate it from FMS described below [39].
SAS and all derived schemes presented below can often be applied to pairs of pump and probe lasers at
different frequencies in very similar configurations. This is useful to lock lasers which are not directly
coupled to a ground state.

5.2.2. Polarization spectroscopy lock

Of the methods derived from SAS, polarization spectroscopy was introduced by Wieman and Hänsch in
1976 [107]. It can be understood as a technique similar to the Hänsch–Couillaud scheme using an induced
polarization birefringence inside a vapor sample instead of a cavity. The setup as shown in figure 6(a)
consists of an SAS configuration which is modified using a circular polarized pump beam combined with
polarization dependent separation and differential detection of the probe beam in transmission. A much
stronger pump beam introduces a birefringence inside the atomic medium due to an anisotropic population
of m states from pumping with a circular polarization [108]. This leads to an ellipticity of the probe beam
behind the vapor sample which changes sign at the resonance and increases angle and magnitude roughly
proportional with detuning [109].
The difference signal features a steep dispersive shape around resonance that can be tuned by changing
beam powers or diameters and then directly be used as an error signal [108–110]. Many advantages and
drawbacks are shared with the Hänsch–Couillaud scheme: common mode intensity noise on the laser beam
is rejected between the two photodiodes and a minimal absorption in a thin or low-pressure sample which is

12
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Figure 6. Two modulation free SAS type locking schemes using counter-propagating pump and probe beams to induce an
anisotropy in the atomic medium. (a) The polarization spectroscopy lock induces a birefringence in the atomic media using a
circular polarized pump. (b) The DAVLL type lock uses a magnetic field to induce a dichroism due to Zeeman shifts.

nevertheless accompanied by a strong dispersion can lead to a sufficient SNR [3, 111]. No modulation of the
laser beam is required, and a larger capture range compared to SAS dither spectroscopy is usually possible
[108], thus featuring a remarkable performance given its simplicity. A side effect of the underlying
anisotropic pumping is that cycling transitions (where the excited state cannot decay into a different state)
are strongly enhanced over non-cycling transitions and cross-over resonances [11].
The technique also suffers from all drawbacks it shares with the other SAS based and non-modulated
locking variants: the capture range is inherently limited by the linewidth of the interrogated transition.
Moreover, the locking point is very sensitive to external fields [3] and the actual atomic resonance is usually
not located at the center of the dispersive slope due to a strong Doppler background [108]. Some of these
issues can be circumvented by sufficient shielding or a slightly modified detection process [112–114] such
that polarization spectroscopy locks are successfully applied in various recent publications [11, 115, 116].

5.2.3. DAVLL
Another approach to introduce an anisotropy into a vapor sample relies on Zeeman shifts through externally
applied magnetic fields. Different schemes which employ modulated and static magnetic fields have been
proposed [117–119], of which the DAVLL remains the seemingly most widespread variant since its
publication in 1998 [120] and extension to SAS around 2003 [121, 122].
Figure 6(b) depicts the principle setup, which, being another SAS based method, shows noticeable
similarities to the polarization spectroscopy lock. Both indeed rely on a balanced detection, but DAVLL is
typically understood to be probing the dichroic σ ± components using a quarter wave plate and PBS
combination instead of the ellipticity birefringent measurement described above. Since the Zeeman splitting
lifts the degeneracy of different m states, the method effectively probes the difference in transmission
between all ∆m = −1 and ∆m = +1 transitions. On resonance, both are identical, while any detuning will
lead to a stronger contribution in either direction. The resulting difference signal is again dispersive and
might be optimized using the magnetic field strength [123].
The DAVLL scheme again shares many of the advantages and drawbacks with polarization spectroscopy
locking, including its simplicity, the lack of modulation at cost of high technical noise, limited capture range,
Doppler background, and sensitivity to external fields. The stability and thermal influence of any magnetic
field coil can be challenging [124, 125], but some effort went into optimizing, embedding, or miniaturizing
DAVLL based locking systems in recent years [4, 126–128].

5.2.4. FMS lock

The idea to lift the error signal generation far away from the low frequency floor of technical noise was not
only explored in the context of PDH with optical cavities, but also as FMS of atomic media throughout the
1980s [129–131] (based on initial work from the 1960s [68]). Similar to the PDH technique, the probe beam
gets phase modulated to generate sidebands ν ± fmod which heterodynely interfere and form a modulated
signal on a photodetector (shown in figure 7(a) including an optional pump beam to create sub-Doppler
features). This contains information about both the absorption and dispersion, the latter of which is
demodulated using a mixer to create a literally ‘dispersive’ error signal. By tuning the modulation frequency
and modulation index with respect to the natural linewidth of the transition, the shape of this signal can be
tweaked for specific locking (or measurement) purposes [83, 131, 132]. As FMS relies on the interference of
the first order sidebands shown in figure 3(b) we repeat our warning against overdriving the modulation
with the intention of a better signal-to-noise ratio. Inheriting the drawbacks and advantages already
described for PDH type lock, FMS suffers from additional asymmetries and offsets arising from nearby

13
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Figure 7. Schematics of two common methods to create an error signal using RF modulation and an atomic vapor. (a) In the FMS
scheme, the probe beam gets directly modulated, such that the dispersion of the sample can be extracted by a frequency mixer. (b)
In the MTS scheme, the pump gets modulated leading to a transfer of this modulation to the probe beam.

transitions [133], which do not exist in a perfectly periodic optical cavity. Being an RF modulation technique
it nevertheless lowers the overall noise floor and reduces offset drifts [134] and can therefore be found in
various recent experiments and atomic species [24, 134–137].

5.2.5. MTS lock

MTS is closely related to FMS to the point where it can be nontrivial to identify either scheme in specific
cases, while even mixed variants have been described [134]. The spectroscopic technique was first described
in the early 1980s [138–141] and applied to locking not later than 1995 [142] ([143] briefly outlines the
history). Figure 7 shows an MTS locking setup, which is remarkably similar to FMS locking, yet different in
that the pump beam is modulated in lieu of the probe beam. The eponymous ‘transfer’ of modulation from
the pump beam to the probe beam ultimately leads to a modulation on the photodetector that, again, can be
demodulated creating a dispersive error signal. Due to the transfer of modulation, the underlying physical
mechanism is a four-wave mixing process in which two probe photons at ν and one pump photon at
ν ± fmod form a signal photon at ν ∓ fmod . These photons then interfere with the probe photons and induce a
modulation at fmod on the photodetector.
In difference to FMS the resulting error signal has a flat, ‘nulled’ background where the resonance peaks
coincide with the zero crossings [141]. This is not affected by laser intensity, laser polarization, and
temperature in first order [143]. MTS, in contrast to FMS, is usually only applied with modulation
frequencies smaller than the natural linewidth, since this produces the steepest, most sensitive slope in the
error signal [141, 144–146] but limits the possible feedback bandwidth. Tuning this steepness might be a
desired property which can be influenced by geometric beam parameters, modulation index, and
polarization [145, 147]. Similar to polarization spectroscopy, only cycling transitions create a significant
MTS signal while crossover resonances are suppressed [147]. RAM can be introduced by MTS itself and
adversely affects its locking performance if present [148].

5.3. Locking to a reference laser

It is a natural question to ask whether it is possible to use an already stabilized laser as an optical oscillation
‘reference’ for another laser, called ‘target’. This is particularly plausible in applications which require
multiple lasers with possibly <10 GHz frequency difference that interact with the same atomic species.
Doppler cooling of atoms using a repumping laser typically fulfills these criteria. There are even
constellations like off-resonant Raman spectroscopy, where only the difference frequency between two lasers
must be stabilized while the exact absolute frequency of both components is less critical.

5.3.1. Transfer cavity locks

Based on the discussions of cavity based locking in section 5.1, a ‘transfer’ locking scheme can be directly
derived: Couple the reference laser to a cavity that can be length-tuned using a piezoelectric actuator and
employ any of the aforementioned techniques to lock the cavity length to a laser resonance. This requires
slightly different feedback loops, since piezoelectric actuators fundamentally lack any tuning faster than their
mechanical properties in the kHz range. The locked cavity can then be used as a stabilized frequency
reference for the target laser, even if e.g. the temperature changes or slow vibrations are applied [149–152]. A
drawback of this method lies in the chaining of individual locks, each of which potentially introduces noise
into the system. It however can be built much more compact than a ULE cavity inside a vacuum chamber,

14
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Figure 8. (a) Schematic drawing of a scanning transfer cavity stabilizing a target laser to a reference laser by observing the
transmission of both triggered by the piezoelectric modulation of the cavity length. (b) The signals captured by the triggered
acquisition show the periodic transmission peaks of the cavity, which are reflected about each typically triangular signal ramp
starting at tstart and ending at treverse . The feedback loop acts on the time difference between ttarget and tref .

and has the opportunity to be virtually infinitely expanded using just one reference laser. If the cavity is kept
under atmospheric pressure, differential refractive index changes between the two wavelengths must be
corrected [25]. The overlapping and separation of reference and target beam can be achieved based on
polarization or wavelength (using e.g. a grating or dichroic mirror). A specific transfer cavity system as used
in our laboratory is discussed in section 6.
Another related and in some sense even more simplified approach relies on continuously modulating the
cavity length (called scanning) using the piezoelectric actuator on a Hz to kHz timescale [153–155]. In this
scanning transfer cavity scheme as depicted in figure 8(a) it becomes possible to observe a low finesse
transfer cavity in transmission and detect both target and reference beam triggered against the scanning
ramp generator. The captured signals (figure 8(b)) will exhibit a certain time difference ttarget − tref that
encodes the frequency difference between both lasers. By keeping this temporal distance constant, again
correcting for refractive index changes, the frequency of the target laser can be stabilized to any desired cavity
resonance. The scanning transfer technique is strictly bandwidth limited by the mechanical action of the
piezoelectric actuator, even if it is driven at mechanical resonance [156]. This softens the requirements
towards the control electronics and allows the implementation of affordable PC or microcontroller based
locking system, which evaluate the signals captured by the photodiodes [25, 157, 158]. It also avoids locking
the cavity length as an intermediate step. Overall, based on the low bandwidth and long delays, the scanning
transfer cavity technique is not employed where linewidth narrowing is desired, but only, when an absolute
frequency stability with the free-running linewidth of the laser is already sufficient.

5.3.2. OFLLs
OFLL and the closely related OPLL, both of which are sometimes referred to as ‘offset locking’ or ‘beat note
locking’, stabilize a target laser at an RF scale frequency offset to a reference laser. Historically, the OPLL as
derived from electronic phase lock loops (PLL) was explored much earlier in the 1960s [159, 160] compared
to the first experiments with OFLLs in the 1990s [161, 162]. In both cases, the two lasers are brought to
interference on a photodetector, creating a beat note oscillation of which only the low frequency difference
|νreference − νtarget | will be directly electronically captured (requiring a fast enough photodetector). This
frequency component is then fed through a FVC and compared to a reference voltage Vref (figure 9(a)). To
increase the accessible RF frequency range, a (variable) frequency scaler, that reduces the frequency by an
integer, is routinely included in such applications. An OFLL error signal often features a wide capture range
that can be employed to lock lasers with high phase noise, low bandwidth feedback loops, and low SNR [163,
164]. This is advantageous if the lock can track a fast sweep of the reference voltage that leads to a fast and
unambiguous sweep of the target laser frequency [116, 165]. Various other architectures based on delay lines
or RF filters in combination with an RF oscillator, which replace the scaler and FVC discriminator, have been
proposed in the literature to overcome their limited frequency range and enhance the overall OFLL
performance [166–170].

5.3.3. OPLLs
Certain types of experiments, e.g. atom interferometry, require phase coherence between multiple lasers. If
the phase between two lasers stays perfectly synchronized—even if their frequencies are at an offset—their
frequencies will also stay in perfect distance. The principle OPLL setup in figure 9(b) illustrates the
underlying idea to compare the beat note frequency with a known stable RF reference oscillator. Compared
to an OFLL, such an OPLL demands a much higher feedback bandwidth to keep the phase slip within the

15
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Figure 9. Schematics of two locking schemes derived from the beat note between two lasers on a photodetector. (a) In an OFLL
this beat oscillation is then scaled (divided by an integer) and converted to a voltage by an FVC. The error signal is obtained by
subtracting a certain reference voltage Vref . (b) The OPLL mixes the beat oscillation with an RF reference oscillation, that governs
the frequency offset. By comparing frequency and phase in a PFD that is clocked by the same LO, the error signal is generated.

narrow capture range of π/2 rad at all times (beyond which the mixer becomes periodic) [37, 169]. PFDs
improve the capture range by comparing both phase and frequency and pulling the laser towards the OPLL
capture range, if the phase difference gets too large. Choosing an exclusive phase comparison over a PFD,
i.e. by low-pass filtering directly behind the mixer, is possible, but only feasible for laser systems with
intrinsically narrow linewidths and high feedback loop bandwidth [160]. In the context of this tutorial, the
OPLL typically requires the highest feedback bandwidth of all systems. Consequently, a wide variety of
publications aiming to soften the requirements, reducing cost [171–173], or shifting the work to an all-digital
PFD exist [174–176].

5.3.4. OFCs
An OFC, if used as a reference laser system, can solve the absolute frequency issue. Its optical oscillation
modes are completely determined by two RF oscillators, which can in turn be locked to an atomic clock. Yet,
it would be an understatement to just call it another laser system, considering its deeply connected history
with frequency metrology, laser locking, and pulsed lasers as outlined by Hänsch in his Nobel Prize lecture
(received for the development of OFCs and contributions to precision laser spectroscopy) [177].
Femtosecond OFCs rely on the fundamental—yet technically not easily exploited—Fourier relationship
between temporal and frequency components of a mode-locked pulsed laser. Mode-locking inside a
non-linear cavity leads to a tight temporal relationship between neighboring frequency modes, enabling to
the emission of a train of short pulses [178–180]. If the emitted spectrum is sufficiently broad and the
dispersion adequately controlled, the pulses with repetition rate frep necessarily yield a harmonic frequency
comb with a longitudinal mode spacing frep , that can be controlled through the cavity length. The frequency
of the Nth longitudinal mode is governed by the ‘comb equation’

νN = N frep + f0 , (11)

where f 0 is an additional, randomly drifting offset frequency generated by a phase shift between successive
pulses. Measuring and fixing frep is straight forward by cavity length tuning, while accessing f 0 is the
innovation that enables building a truly self-referencing OFC. If the spectral bandwidth of the comb spans
more than an octave—that means two frequency components exist where some ν2N = 2N frep + f0 have at
least twice the frequency of some other ν N — this lower frequency can be doubled and brought to
interference with ν2N to measure the beat note 2νN − ν2N = f0 . Now, an intra-cavity dispersive optical
element changes the group delay and stabilizes f 0 , thus fixing all ν N precisely based on two RF frequency
measurements.
Locking a laser to an OFC is often performed using an OFLL or an OPLL, which additionally selects a
single frequency component of the comb (often the closest) by means of an electrical band-pass filter [181].
The ambiguity or discontinuity when tuning the laser frequency across multiple comb frequencies can be
overcome, such that fast and wide (many nm) frequency sweeps locked to an OFC have been demonstrated
[182–185]. Interest in frequency locking with OFCs is particularly strong in precision metrology, due to their
absolute frequency standard and possible narrow-line phase locks [186, 187]. For applications with cooled
atoms or hot atomic vapors, an OFC can form the shared single source of stability for many laser systems at a
range of wavelengths [163, 188–190]. This can offset the high upfront cost that is associated with an OFC, yet
depending on the spectral range multiple OFCs might be necessary, if they are available at all.

16
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Figure 10. (a) An OFC generates a series of well-defined optical frequencies which are fully determined based on two electrical RF
measurements. Each of those frequencies can subsequently be used in a OPLL or OFLL like drawn here. (b) A wavemeter can
directly measure the exact frequency of the laser based on one or more internal interferometers. For high absolute precision this
requires regular calibration based on a known stable light source.

5.4. Locking to an interferometer or wavemeter

Optical wavemeters are routinely used with single frequency lasers to set and monitor wavelengths. This is
particularly true in combination with any of the presented frequency discrimination techniques, e.g. to verify
that the correct mode of a cavity or an OFC has been selected. These devices are really interferometers of
different kinds, such as scanning Michelson interferometers [191, 192], multiple Fabry–Pérot
interferometers [193], and multiple Fizeau interferometers [194–196] in combination with fast acquisition,
signal processing, and sufficient calibration.
Since wavemeters with a relative precision of 10−8 and better (corresponding to less than 10 MHz
absolute precision in the visible spectrum) became commercially available [197, 198], they have been used to
stabilize the frequency of lasers [199]. Wavemeters offering this precision require frequent calibration and are
commonly operated in combination with a known stable reference light source (figure 10(b)). The biggest
advantage of wavemeters lies in their virtually infinite capture range, that is far wider than the emission band
of most laser sources, and the possibility to lock at any frequency within this range without additional
components or modulation. Due to the signal processing involved, wavemeters typically have acquisition
rates of less than 1 kHz, and therefore offer only limited feedback bandwidth, essentially ruling out any
linewidth narrowing. As high precision wavemeters also come at increased cost, it is common to use them in
combination with fast optical switches and multiple target lasers simultaneously [200, 201]. This limits the
acquisition rate for each individual target laser even more, but does not strongly impact the locking
performance, as only a suppression of frequency drifts is possible anyway. Research groups have successfully
demonstrated the use of such systems for Doppler cooling and trapping of various atomic species [202, 203].

5.5. Common extensions and other types of frequency locks

Complex experiments often rely on different laser locks for different systems, e.g. a wavemeter lock for a
series of broad ground state transitions, supplemented by narrow linewidth OFC locks or even extremely
narrow PDH cavity locks. Locking a single laser using multiple of the presented frequency discrimination
schemes is necessary, e.g. if two seemingly incompatible requirements arise, i.e. locking a noisy laser to a
narrow linewidth cavity but with outstandingly large capture range or stability. Some techniques like PDH
and FMS require an RF modulation, but can only be applied directly at a resonance. Sideband locking
overcomes this limitation by modulating most of the laser energy into a sideband at the desired offset and
locking to this sideband [204]. Finally, electronic feedback might be avoided completely in favor of purely
optical feedback beyond the internal laser cavity, e.g. from an external cavity (as done in ECDLs), from an
atomic transition [205], or injection locked from another laser or OFC [183, 206, 207].

6. An exemplary locking scheme

Given the wide range of possible applications with hot atomic vapors alone, it is impossible to give universal
advise on designing a laser lock in the style of a generally valid recipe. In this section we will much rather
study one system that was developed and extended in our laboratories during the past years. We will focus
particularly on the process that has led us to settle with and eventually build the setup as it stands, while
covering some more general aspects that were omitted in the previous sections.

17
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

6.1. Decision process

Our research activities using thermal atomic vapors include vapor-cell manufacturing and testing, EIT type
schemes involving the Rydberg blockade, nanosecond four-wave mixing in microscopic vapor-cells, gas
sensing applications, and integration of photonic chips with thermal atomic vapors. Currently, cesium
atoms, rubidium atoms, and nitric-oxide molecules are the primary subject of research, but other atomic and
molecular species have been used or might be used in the future. In contrast to many experiments involving
cooled or trapped atoms that require specifically designed vacuum chambers, Zeeman slowers, or magnetic
field coils, it is feasible to perform studies involving different atomic species (e.g. Cs and Rb) and switching
between those in ‘quick’ succession by exchanging vapor-cells. To facilitate such research, we operate a set of
shared lasers, e.g. the alkali D-line transitions and their successive Rydberg transitions, which are distributed
from a central, rarely entered, temperature stabilized room using polarization maintaining fibers. These are
supplemented by laser systems more specific or tailored to particular setups located close to their intended
application, such as pulsed amplifiers, unusual excitation paths, or high power beams.
The setup as previously operated was highly heterogeneous, containing numerous self-made ECDLs,
commercial ECDLs, and SHG cavities locked through a mixture of commercial and self-made PID
controllers. Optical frequency discrimination was performed primarily against atomic standards using
DAVLL and dither locking but also using PDH with a ULE cavity. Since the students and researchers, who
had originally conceived those laser locks, mostly had left for different laboratories and the electronics
increasingly showed aging-related failures, a replacement was sought for the next generation of laser systems.
Our requirements and informal wish-list for laser locking therefore included:

• Direct locking of at least 12 lasers at a wide range of wavelengths between 420 nm (Rb 5S1/2 → 6P1/2 ) and
1529 nm (Rb 5P3/2 → 4D5/2 ), including multiple lasers at identical or almost identical wavelengths.
• Linewidth of the lasers should be on the order of few 100 kHz (few kHz linewidth might be necessary in the
future).
• Locking is possible with large, arbitrary detuning from atomic transitions (some planned excitation schemes
require hundreds of GHz detuning from atomic resonance, others use weak transitions between high NO
molecular states).
• Ideally, a single solution exists for all laser locks.
• Easy and cost-effective extensibility and scalability for additional laser systems.
• Electronics are readily available for purchase or reproduction (and can ideally be modified for other feedback
applications such as intensity stabilization).
• Monitoring and remote control possible via our institute’s Ethernet network during experimentation in
neighboring rooms.
• Digital electronics are generally favorable since locking settings can be stored and reproduced.
• Ideally open-source software gets employed for most tasks.
• Possibility to continue using all existing laser systems.
• Achieves a homogeneous way to introduce new students to the locking system.

Locking lasers at arbitrary frequencies essentially rules out any atomic vapor reference, which would have
been only possible in some of our use-cases. A wavemeter or OFC offers a reasonable turn-key solution, but
turned out to require multiple of the most-expensive devices commercially available. It also lacks
cost-effective gradual scalability as new wavelengths or a larger number of lasers might require additional,
expensive hardware.
We therefore decided to design a cavity based system, which could additionally serve as a frequency ruler
during scanned operation of lasers. A ULE cavity with an FSR of roughly 1.5 GHz and a finesse of about 3000
at certain wavelengths for rubidium spectroscopy was already available. The purchase of additional ULE
cavities to be used with a set of lasers overlapped through dichroic mirrors was considered but impractical
due to high upfront cost, complexity, and limited extensibility, if e.g. multiple laser need to be operated at
adjacent frequencies.
Eventually the decision came towards transfer-cavity locking, with a reference laser either locked to a Rb
5S1/2 → 6P3/2 hyperfine ground state transition using DAVLL or to a ULE cavity using PDH. The former
leaves the ULE cavity unoccupied for other applications, but makes transfer to another laser at the same
atomic transition impossible with a dichroic mirror, while the latter permits linewidth narrowing.
Modulation on the reference laser can be introduced through current modulation, as it does not serve any
purpose beyond stability transfer. Stabilizing the target lasers might either be done by locking the transfer
cavity length using the same modulation and PDH, if a suitable modulator is available for the target laser,
too. If not, the same cavity can be operated as scanning transfer system. The standardized optical setup

18
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Figure 11. A schematic depiction of the transfer cavity setup as used in our laboratory (detailed description in the main text). In
the leftmost box a reference laser gets locked to a ULE cavity or a DAVLL configuration using PDH. Each box to the right indicates
an additional laser getting locked to the reference laser by means of a transfer cavity (more such systems can be added as
required). The signals are controlled by means of RedPitaya STEMlab 125-14 (RP) modules which are daisy-chained to share the
same clock signal to allow for demodulation of the shared reference oscillation on the reference laser.

around self-built cavities (with about 1 GHz FSR and a medium finesse of about 1000) is clearly and logically
separated from the feedback electronics, which can be upgraded or exchanged if required.
A drawback of this architecture lies in the large number of frequency locks and the spatial separation of
up to 30 m between the reference laser and some specific laser systems in other rooms. While it is possible to
locate the transfer cavity on either end, RF signals must be transferred for some PDH lock and feedback over
this distance. Additional care must be taken to avoid any form of cross-talk between the different locks in
both the optical and electronic domain. Our colleagues in a neighboring lab took a similar approach based
on scanning transfer cavities as described in [208].

6.2. Setup
We show a sketch of our setup in figure 11. Here, we divided it logically into the reference lock and two
subsequent transfer locks ‘A’ and ‘B’. Each transfer lock can be replicated such that it meets the amounts of
locks needed. The lower part of the figure shows the required electronics, which is based on the use of several
STEMlab modules and discussed later. We discuss the optical setup first.
The main reference laser is either stabilized to a ULE cavity or a (rubidium) vapor cell. A choice between
both possibilities is made by using the combination of a half–wave plate ‘λ/2’ and a PBS. In our setup this
laser operates at 780 nm. The stabilization to the ULE cavity is accomplished by employing the introduced
PDH locking technique. The laser is coupled into the ULE and the transmission monitored by a PD labeled
‘TULE ’. A quarter waveplate (λ/4) and another PBS guides the reflected light to another PD labeled ‘RULE ’. It is
possible to use DAVLL as an absolute frequency reference as seen in figure 6, though we currently do not
make use of this possibility. Still, this is sketched in figure 11.
Each transfer lock is optically designed the same way. Our transfer cavities are self-built [209], roughly
15 cm in length, and have an FSR of about 900 MHz. Their mirrors are coated such, that they reflect both the
wavelength of the reference laser and the wavelength of the target laser. The reference laser is used to stabilize

19
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

the length of the transfer cavities. As this can only be done as soon as the reference laser itself is locked to the
ULE, we are not able to scan the laser anymore to observe the PDH signal. Thus, we equipped each transfer
cavity with a piezo attached to one the mirrors, such that it is possible to modulate the transfer cavity’s
length. The piezo’s voltage input is labeled as VPiezo . We use a dichroic mirror to overlap both beam paths. A
BS behind the transfer cavity equally splits the passing light between two PDs labeled as T 1 and T 2 , and used
for monitoring the transmission. By using a bandwidth filter the PD either monitors the reference or the
target laser. As the resonance frequency of the transfer cavity is not necessarily equal to the targeted locking
wavelength, we guide let the target laser pass an EOM to introduce sidebands for offset locking. The EOM’s
modulation input is labeled as ‘EOM Mod’. We then monitor and demodulate the reflected light of both laser
beams labeled as ‘R1 ’ and ‘R2 ’, respectively.
The electronic setup is based on several RPs, which are controlled by our slightly modified version of
PyRPL [28, 210]. A single RP board features two inputs and two output ports, and PyRPL implements an
ASG, an IQ demodulation, and a PID as well as a scope running all directly on the RP’s FPGA. For the PDH
locking technique we need a phase-modulated laser. We use a first RP to modulate the current of our
reference laser, in our case at 15 MHz. This corresponds to module 2 in figure 11. On another RP (1 in
figure 11) we monitor the transmission and reflection of the ULE. This RP is also used to generate the PDH
error signal out of the monitored reflection. However, since this is a separate board, the modulation’s board
clock and demodulation’s board clock are not in sync. As such it is impossible to generate a reliable locking
signal, as the signal would drift over time. We adjusted PyRPL such, that the clock source of the very first RP
is daisy-chained to all others by means of SATA–cables. This enables not only the reliable locking of the
reference laser, but also the subsequent locking of the transfer cavities, which is also based on the
demodulation of the reference laser’s signal. For any RP controlling the lock of a target laser such a
modification is not necessary, as long as only a single board can be used. Our reference laser is a Toptica DL
Pro, which enables us to give both fast (current modulation) and slow (piezo tuning) feedback to our laser
using PIDs, as labeled accordingly in figure 11. For a transfer lock, i.e. a lock stabilizing the length of a
transfer cavity, another RP (3 in figure 11) is used, connected to the piezo via an output port, and
monitoring the transmission and reflection of the two respective PDs. Here, we only use a single PID for the
stabilization. For any target laser we add an EOM, which is in our case wired to a Windfreak SynthUSBII as
the RF source for offset locking. Other than that the wiring is identical to that of the reference laser, an
exemplary shown with RP 4 in the figure. Depending on the target laser’s type either a single or more PIDs
are used. Next, we will focus on the typical locking procedure.

6.3. The locking process

In principle, we could span several UIs provided by PyRPL for each RP. However, to ease the usability and
further implement a ‘lock watch’, we wrote a wrapper around PyRPL, which essentially spans a separate
PyRPL process per RP, but combines them all in a single interface. Screenshots can be found in figures 12
and 13. As any transfer cavity lock relies on the lock of the reference laser, and any target laser lock relies on
the lock of the transfer cavity, we always have to lock in order.
We start by locking the reference laser. At first, we turn on the current modulation ‘Cur. Mod.’ of the
reference laser by using the ASG of the second RP. Then we use an ASG of the first RP to apply a ramp to the
reference laser’s slow feedback port ‘[Link] .
slow ’ which controls a piezo driven Littrow grating. On the RP’s
scope we observe the 00–mode transmission peaks ‘TULE ’ of the ULE as well as the internal ramp signal, and
are able to center such a peak on the ramp by adjusting the ramp’s offset. We continue this process whilst
lowering the ramp’s amplitude. When the amplitude is low enough that only a single transmission peak is
clearly visible, we change the output to a constant DC voltage. Next, we repeat the same process with the fast
feedback port ‘[Link] .
fast ’ which modulates the diode laser’s current and another ASG module of the same
RP. As soon as the signal is sufficiently centered we take a look at the PDH error signal generated by the IQ
module of the RP with the input information being the reflected signal as well as the modulation frequency.
Typically, the phase is not correct at first and the PDH signal is weak, yet we are able to correct for this from
within PyRPL. Next, we adjust the setpoint of a PID, which gives feedback to the fast feedback port, such,
that it is in the midst of this error signal. We then slightly increase its ‘P’–part. By observing the transmission
peak, which should broaden if the phase is correct, we are able to determine if we are 180◦ off or not. If the
phase is correct, we continue locking by disabling the ASG and firstly increasing the ‘I’–part and
subsequently the ‘P’–part of the PID. Both are tuned such that they are as large as possible without inducing
an unstable oscillation on the transmission signal. If locking was successful, the transmission signal should
have turned into a flat line roughly at the maximum of the former transmission peak. To increase the stability
of our lock, we use a second PID which gives feedback to the slow feedback port based on the fluctuations of
the first PID. It uses the output voltage of the first PID as error signal and a setpoint of zero which makes sure
that the fast PID has maximal range for rapid adjustments in any direction. The ‘P’ and ‘I’ values are tuned

20
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Figure 12. Screenshot of the graphical user interface implemented by us on top of PyRPL showing an overview screen include the
lock watches.

Figure 13. Screenshot of the graphical user interface implemented by us on top of PyRPL showing the reference laser lock screen
during the locking sequence.

identically to the first PID by slowing increasing their value and observing the onset of unstable oscillations.
If everything went well we are able to turn a ‘lock watch’, which simply checks if the mean of the transmission
signal stays within a certain range.
In a next step we can lock our first transfer cavity. Here, an ASG of the respective RP is used to apply a
ramp to the piezo. Again, we observe the transmission peaks of the reference laser, and are able to decrease
the scan amplitude while keeping the transmission signal centered. The IQ module generates the error signal,
and the PID then gives feedback to the piezo. Here, only a single PID is used.
As a final step, we lock our target laser. In comparison to the lock of the reference laser the only difference
is, that we can also adjust the EOM’s offset frequency. For example, if we would like to lock to a certain
atomic or molecular transition, we make sure that we have both the cavity signal and the transition signal
available, and then tune the frequency such, that the error signal is in line with the transition signal’s peak.
We then continue locking with this signal.

21
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Figure 14. Linewidth measurement using a delayed self-heterodyne interferometer. (a) Schematic setup of the interferometer,
where a portion of the beam gets delayed and another portion is frequency shifted using an AOM. The polarization drift behind a
non-polarization maintaining fiber is compensated using waveplates, such that it matched the polarization of the other beam. (b)
An example of the resulting power spectrum as observed on a spectrum analyzer for one of our Rb 5S1/2 → 5P3/2 transition
lasers at 780 nm showing a reduction of the linewidth. Typically, such plots would also show servo-bumps generated in the
locking feedback loop, which are not resolved above the noise level in this measurement.

6.4. Characterization
We have already introduced fundamental concepts of the characterization of laser frequency stability in
section 2. Here we want to briefly exercise how these theoretical quantities are measured in the lab to verify
and optimize the performance of the overall laser system. In many cases it is beyond the scope of a quantum
optics experiment to quantify all possible frequency stability parameters. In many cases we have no second
laser to perform a comparative beat-note for a proper Allan deviation characterization and in some cases we
cannot even perform a delayed self-heterodyne linewidth measurement due to bad transmission of our
available delay fibers. There are still some tricks to develop trust in the performance of the laser lock.
The transmission and error signals can provide a first idea of the locking performance. A ‘flat’ error
signal is a good indication of a successfully locked laser. However, when observing such in-loop quantities,
one must always remember that they are already contaminated by systematic errors:
The literature is full of ‘outstanding’ results based on improper measurements. A good example
of a very popular mistake is to lock a laser’s frequency [. . .] and then monitor the error signal
within the servo-loop to provide a measure of the residual frequency noise. This appears reas-
onable at first glance, but usually the error signal is contaminated at some level by systematic
errors [. . .]. Any measurement of the error signal would indicate that the laser’s frequency fluc-
tuations are small, because the servo loop is acting to cancel the error signal. [58]
Nevertheless, such an evaluation can be useful if the results are interpreted with caution [111].
We continue by observing long term effects and drifts in the overall system. Long term stability in our
case is primarily limited by disruptive events such as air pressure fluctuations created by moving doors or
strong acoustic noise that couples particularly well to the transfer cavities. Whilst the reference laser coupled
to a ULE cavity under vacuum often stays locked for many days or even weeks without interruption, the less
shielded environment of transfer cavities in air tends to produce a loss of lock for some cavity or laser on a
daily basis. These disruptive events are heuristically observed by monitoring all control signals and assuming
a loss of lock as soon as an unusually fast change is detected. Often it is possible to perform automatic
relocking by setting the relevant control signals back to their last recorded locked state.
Sometimes we can evaluate long term drifts of the locks by comparing them to a known stable reference,
such as an atomic vapor cell or ULE cavity. We confirm that e.g. a 422 nm ECDL laser locked to the Rb
5S1/2 → 6P1/2 transition stays within the natural linewidth of the atomic transition during our experimental
days by observing an SAS signal in a reference vapor cell which is sufficiently stable for our thermal
experiments.
As a next step, we investigate the short term linewidth of our locked lasers using a delayed self-heterodyne
interferometer as shown in figure 14(a) [211, 212]. By frequency shifting a portion of the light with f 0 , the
scheme recovers an autocorrelation of the power spectrum on an electric spectrum analyzer around f 0 . A
longer delay time and thus a longer delay fiber permits the measurement of lasers with narrower linewidths
and longer coherence times. We show an exemplary measurement for our 780 nm reference laser with a
commercially available, 10 km long delay fiber produced for telecommunication wavelengths at
1330–1550 nm. While this leads to high losses inside the fiber, it is still possible to recover a signal above the
detection noise that confirms that we are able to narrow the linewidth of the laser to below 60 kHz. In
practice such results also need careful analysis and interpretation: The autocorrelation does only truly

22
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

capture specific lineshapes. Fiber lasers have different noise source leading to a typically broad Gaussian
noise spectrum. Nevertheless, there are ways to gain information on lasers with a coherence time longer than
the delay time [213].
Beyond that, depending on the necessary performance and system characteristics, a variety of tools are
available to describe and measure system noise, long-term locking stability and frequency resolved laser
lineshapes, which we cannot cover in this tutorial. The noise performance of all electronic components
including a model of the feedback loop electronics might be used to predict fundamental the limits of a laser
lock [37]. The Allan variance as described in section 2 measures how well an oscillator (or more specifically a
laser) performs over an extended period of time as noise continuously averages out [45]. As this is always a
comparative measurement between two oscillators—in this case e.g. lasers or optical resonators—it requires
an independent optical [60]. If no adequately stable, narrow, and uncorrelated reference exists, it is possible
to extract information by comparing two identical but completely independent systems [35]. In the context
of atomic optical clocks, the three-cornered hat method is employed, where three or more different and
uncorrelated systems are compared with each other to measure the observed frequency deviations on all time
scales [214, 215].

7. How to fabricate a vapor cell

The age of modern atomic spectroscopy was initiated by the invention of the low pressure sodium lamp in
1919, when Compton signed a patent in the USA as an employee of Westinghouse Electric [216]. This lamp
was operated at a pressure of a few mbar to maintain enough ionizing collisions for good brightness and
featured a comparably narrow spectrum. As we know from our work a century later, alkali vapor cells tend to
become opaque when operated at elevated temperature. Compton also ran into this problem and developed a
new type of glass [217], which was a very boron rich variation of the usual borosilicate glass invented by Otto
Schott in 1887. Most of the standard spectroscopy cells, especially reference cells, are made of borosilicate
glass to this day, which run under brand names like Duran, Pyrex, Borosil, Borofloat, and many others.
Yet there are many other materials beside borosilicate glass, which can be used to make vapor cells. These
are supplemented by a plethora of fabrication processes beyond the traditional glassblowing. In the following
subsections we will introduce to the most common materials and methods and contextualize their use within
the field of vapor cell fabrication.
A more extensive review for the specific case of MEMS vapor cells can be found in the PhD thesis of
Karlen [218] and the article by Knapkiewicz [219].

7.1. Design considerations

In the first part of this tutorial we introduced vapor cells in the form of reference cells as a tool for locking
lasers. In a broader sense, we want to cover any spectroscopy cell, i.e. defined volume, which holds atoms or
molecules in gaseous state to be probed by lasers, electric and magnetic fields. Depending on the
temperature, we speak of a vapor cell as long as the temperature of the medium stays below the critical
temperature and thus a portion stays liquid or solid. As soon as this critical temperature is surpassed, the
same cell can become a gas cell in which all particles are gaseous.
We can also differentiate such cells depending on the idealized thermodynamic flow of particles:

• Open cells are connected to their environment, e.g. through a series of pipes, which allows exchange of
particles with an external reservoir. This typically requires some form of active pumping or evacuation to
uphold specific flow or pressure conditions.
• Sealed (or closed) cells are disconnected from the environment, thus allowing no further particle exchange.
As we will discuss below, actual implementations might still permeate and desorb atoms or include getter
materials to ensure constant conditions.

There are many applications beyond the laser locking reference, which see the use of spectroscopic cells such
as sensing, atomic clocks, research on electromagnetic interactions (EIT, photon storage, etc), and gas
analysis.
Thus, depending on the intended application, very different requirements can arise. Table 2 lists many of
these criteria which can serve as a starting point during the planning phase of a vapor cell design. In the
following sections we will focus on common materials and fabrication techniques.

7.2. Bulk materials for vapor cells

The most common material found in vapor cell is glass, due to its transparency, vacuum tightness
and—depending on the type—chemical compatibility with alkali vapors. Its most striking feature is the

23
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Table 2. A summary of aspects and constraints to consider during the planning and fabrication of a vapor cell design sorted into
sections. The left column contains key criteria while the right column provides additional context and specific circumstances under
which the criteria become relevant.

Chemical compatibility and atomic medium

Atomic medium • Required amount and purity, method of filling

• Cell materials must be chemically compatible
Additional internal media • Buffer gases, getter materials, chemical reactants
Leak Tightness
Lifetime / Aging • Chemical degradation, deposition, leakage, and contamination
• Reaction with and diffusion into cell surface

Optical, electronic, and magnetic interface

Transmitted wavelengths • Suitable material, polished surfaces, unobstructed optical path

• Might require anti reflection coating, Brewster angled windows
Maximum optical power • Damage threshold and optically induced chemical reactions
Additional optical functions • Photonic structures, waveguides, optical fibers
• Internal deflection (mirror surfaces, prisms, gratings)
Electronic contacts • Applying AC/DC fields, electrical heating, electronic readout
Magnetic shielding • Typically non-magnetic materials to avoid frequency shifts

Environmental conditions

Temperature and Humidity • During operation and fabrication (typically non-condensing)

Pressure • Typically ambient pressure, might be used for length tuning
Surrounding chemistry • Protective atmosphere, laboratory, or rough environment?

Physical parameters

Shape, Volume, Mass • Size of enclosures, ovens, required optical access

Mechanical Strength • External mechanical and thermal loads

Economic constraints

Manufacturability • Single prototypes, many different designs, or industrial scale

• Expected yield, availability, and timescale of fabrication
Cost

Table 3. Typical properties of glass materials used for the fabrication of vapor cells.

Fused Silica Borosilicate

Softening point ≈1600 ◦ C ≈1200 ◦ C

Strain (annealing) temperature ≈1050 ◦ C ≈540 ◦ C
Thermal expansion coefficient ≈5 × 10−7 K−1 –6 × 10−7 K−1 ≈3.3 × 10−7 K−1
Transparency ≈185 nm–3500 nm ≈350 nm–2000 nm

ability to produce high quality, vacuum tight joins by means of conventionally glassblowing. There are
however a range of materials beyond glass that can be used such as silicon, e.g. in parts of the vapor cell that
do not transmit visible light.
Most glass-blown cells are made of hard glass with a high softening point and low thermal expansion
coefficients below 60 × 10−7 K−1 like borosilicate glass or fused silica (amorphous quartz) [220]. Fused silica
(SiO2 ) typically has better optical properties than borosilicate and exists in a large variety of purities. Well
known manufacturers are Heraeus (Herasil, Suprasil, Homosil, Infrasil) and Corning (HPFS High Purity
Fused Silica 7979, 7980 and 8655) among many others. Borosilicate only contains 70%–80% SiO2 . Other
ingredients are B2 O3 , Na2 O, K2 O, Al2 O3 , and other oxides. The alkali contribution is particularly important
for anodic bonding as described below. The main manufacturers are Corning (Pyrex) and DWK Life Science,
formerly Schott (Duran). Typical properties of fused silica and borosilicate glass are summarized in table 3.
Soft glasses such as soda-lime glass have an even higher alkali fraction and are typically of lower optical
quality. Their lower softening point reduces thermal and mechanical shock resistance. Many optical glasses
like crown glass and flint glass are not well suited for glassblowing.
Beyond glass, the fabrication and joining of the vapor cell typically becomes more complicated as we will
discuss in the sections below. Sapphire (Al2 O3 ) comes to mind as a well known anti-diffusion coating in its
amorphous form (called alumina). Bulk sapphire alkali vapor cells have been successfully made and are used

24
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

for research [221] involving high temperature at which glass-made alkali vapor cells already exhibit excessive
diffusion [222].
Silicon (Si, and to a lesser extent also related materials like silicon nitride Si3 N4 and silicon carbide SiC)
forms the basis of many contemporary activities in the fields of nanophotonics and MEMS vapor cells as it
allows integration of functional devices into vapor cells. The advanced capabilities and methods of the
semiconductor industry allow for complex nanostructuring and integration which lead to the
miniaturization of vapor cell based devices such as atomic clocks or magnetometers. Silicon itself is
compatible with most alkali metals, partly transparent for infrared light, and vacuum tight in its crystalline
form.
Other notable material candidates for vapor cell fabrication include crystalline quartz, ceramic materials,
metals, and potentially even plastics. Metals play a particularly important role in the context of anodic
bonding as discussed in the next section and form the basis of most open gas cell vacuum systems.
Due to the strong influence of buffer gas pressures and amount of the atomic medium in the context of
frequency reference vapor cells, much attention has been spent on the interaction between these materials
and alkali vapors [218]. It turns out that both fused silica, and borosilicate glass have a good resistance
against alkali vapors, while soft glasses tend to be easily reduced. The reactivity of lithium in combination
with the rather high melting point of 180.5 ◦ C stands out here and does not allow for a stable operation of a
vapor cell as we discuss here. Therefore, the vast majority of cells are filled with rubidium and cesium, and in
some occasions with sodium and potassium. If one wants to work with lithium vapors it is indispensable to
return to the very well established techniques of discharge cells [223] and heat pipe ovens [224], which are
used up to date. Compatibility of metals with alkalis has been investigated in terms of corrosion and
solubility of the alkali in the metal [218].
It is important to know that one cannot easily combine each pair of materials due to their different
thermal expansion coefficients. A temperature change of a hybrid cell of e.g. fused silica and borosilicate glass
leads to immediate destruction from internal stress already in the production state. If multiple materials
must be combined, they either need to have matching coefficients of thermal expansion or a smooth
transition from one to the other must be created. In the case of the fused silica to borosilicate glass transition,
adding layers with different alkali concentration transitions the thermal expansion coefficient smoothly as it
is done for glass metal transitions in vacuum technology.

7.3. Functional internal coatings and materials

While the outer walls of a vapor cell primarily need to provide a vacuum tight enclosure for the atomic or
molecular medium, a wider range of functional materials and parts can be integrated inside those
boundaries. These include e.g. optical fibers, mirrors, surface coatings, getter materials, and buffer gases. The
available materials are still restricted by the chemical comparability but might also protect the cell walls.
Anti-relaxation coatings have been used as early as the late 1950s [225] and are used up to this day to
increase the spin-polarization lifetime in atomic media [226]. The most common material in this context is
paraffin with a number of alternatives to be used e.g. at elevated temperatures.
Anti-diffusion coatings decrease the diffusion rate of the atomic medium into the vapor cell walls. We use
dense alumina (Al2 O3 ) layers made from ALD of about 10 nm thickness to prevent cesium and rubidium
from attacking glass surfaces made of fused silica or borosilicate glass [227].
Vapor cells with thin-film anti-reflection coatings are commercially available. Similarly, it is possible to
create reflective thin-film reflectors inside vapor cells [228]. These coatings should only employ materials
that inherently resist the atomic media or might be closed with an anti-diffusion layer.
Unwanted gases that reside inside the vapor cell can lead to adverse effects such as pressure broadening or
decreased spin-relaxation lifetimes. While this can be actively minimized by a vacuum bake-out and
sufficient pumping during the filling stage, a closed vapor will eventually leak, outgas, or form gases from
internal chemical processes. This is counteracted by the inclusion of so-called getter materials, which react
with the vapor cell atmosphere and binds any unwanted trace gases such as H2 O or O2 [229].
Finally, buffer gases are purposely added to vapor cells for various reasons. In alkali vapor frequency
standards, buffer gases are added to narrow hyperfine transitions below the normal Doppler width by
increasing the effective lifetime inside the laser beam [230] (similar to anti-relaxation coatings).

7.4. Available fabrication processes for sealed vapor cells

All aforementioned materials can be processed using a wide range of techniques including conventional
milling, fine optical processing, micro-structuring, and scientific glassblowing. The eventual challenge is the
production of a durable and vacuum tight seal. We can distinguish between two main methods for vacuum
tight cell joining: scientific glassblowing and mechanic-chemical bonding. One could potentially even

25
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Figure 15. Steps to make a reference cell using conventional glassblowing techniques (details in the text).

Figure 16. Cutting the glass tube to the desired length (see figure 15(b)).

produce a sealed metal based vapor cell using conventional vacuum flanges and components. Maintaining a
low enough pressure will however typically require active pumping in such cases.

7.4.1. Cells joined by scientific glassblowing

Figure 15 outlines a typical progression during the fabrication of a reference vapor cell. While these steps are
usually performed by a skilled scientific glassblower [231], we will highlight some important steps to
consider during a scientific planning phase. In the first step a borosilicate glass tube of 30 mm diameter with
1.5 mm wall thickness gets cut to the desired length by developing pulling points on both ends. The left side
in figure 15(a) is fully formed into a test-tube like end, while the right side is kept as a holder to handle the
part in the following steps. A T-seal joint towards a tube of smaller diameter is prepared which is going to
serve as access point for evacuation and filling. In the next step, the ends are cut off using a diamond glass
cutting disc as shown in figure 16 under continuous water cooling.
The windows which get prepared in figure 15(c) are polished to the desired optical quality and have
approximately the same thickness as the tube’s walls. The rough surfaces on the tube’s face and the window’s
edges get fire-polished to prevent cracking and ensure a leak-free joint. The windows and the tube get joined
using a glassblowers lathe in which the tube is held in a three-jaw chuck on one end, while the window is
hold using a vacuum chuck.
While glass is only a weak heat conductor and allows e.g. to hold the part during the fabrication, the
required melting temperatures typically prevent the presence of most coatings during the joining process. A
slight variation of purely glass-blown cells is the method of glass frit bonding [232]. Here, the individual

26
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

parts are connected by a lower melting point glass powder or paste as an intermediate bonding layer. The
assembly with the frit applied to the interfaces is then placed in an oven at a set temperature below 450 ◦ C
and pressure where the frit locally melts without the remaining cell. Frits of suitable thermal expansion exist
for almost any material, especially for borosilicate, quartz, and even sapphire. Most commercially available,
macroscopic reference alkali vapor cells are either made by scientific glassblowing or frit bonding.

7.4.2. Anodic bonded cells

Anodic bonding [233] is one of the most common methods to produce MEMS type cells, especially on an
industrial scale. Typically, two borosilicate windows are anodically bonded to a silicon spacer. By applying
heat (up to a few 100 ◦ C) and high voltage (up to a few 100 V) across the stack, ions in the borosilicate
(mostly sodium) start to diffuse away from the silicon while oxygen diffuses towards the silicon. The
negatively charged oxygen ions form a covalent bond with the silicon. Its bond strength can be improved by
adding a thin layer of aluminum, which leads to a connecting Al2 O3 layer. The first cell based on two
borosilicate windows bonded to a silicon spacer has been realized in 2004 [234]. Most commercial
magnetometers and atom clocks us cells made by anodic bonding. For other applications it is also possible to
include vacuum compatible electric feedthroughs [235], e.g. to apply electric fields in a controlled manner or
for direct current readout of highly excited Rydberg states [236].

7.4.3. Other methods to join alkali vapor cells

Besides glass-blown cells and cells made by anodic bonding many other methods have been developed to
fabricate a vacuum tight, long-lived spectroscopy cells for alkali atoms. The simplest approach is to use a
vacuum tight glue. In most cases this is not a suitable solution as these cells cannot be operated at elevated
temperatures, due to the differing thermal expansion coefficients, and a reduced lifetime of a few weeks to
months. It is not clear, whether the cells develop leaks, the glue outgasses or chemical reactions of the alkali
atoms reduce the lifetime. A typical observation is, that a background pressure develops over time and
pressure broadening prohibits spectroscopy at some point. This process is proportional with temperature.
Cells sealed with vacuum epoxy like Torr Seal can last many months when operated at room temperature.
One exception to this maxim are the cells produced in Armenia by the group of D Sarkisyan. Their sapphire
cells are joined using a mineral high-temperature solder [237] adapted from the fabrication of sodium vapor
lamps which belong to the class of eutectic 49% Al2 O3 , 45% CaO, 6% MgO compositions with added ZrO2
[238, 239].
A less common method resulting in robust vacuum tight vapor cells is thermocompression [240]. Here, a
copper layer is added to the two pieces to be connected and then pressed onto each other at elevated
temperature. In [240] they used 500 nm of copper which was fused at 350 ◦ C–400 ◦ C and a pressure of
14 kN. The main advantage of this method is that other materials beyond borosilicate can be used for the cell
as the mobility of ions in borosilicate (anodic bonding) is not necessary. For cells with very small volumes
this method has the additional benefit of reduced residual outgassing compared to anodic bonding.
Although sealing vapor cells by local heating with a laser beam seems to be an obvious method, not too
many attempts along this line have been carried out. Researchers mostly used a CO2 laser to apply localized
heating at a certain position to perform effectively glassblowing on a microscopic scale [241, 242]. The
possibility to apply the laser beam only at a specific spot, leaving the rest of the cell at low temperature, has
been exploited to make a MEMS type cell with a local breakseal [243].
The last glass joining method mentioned here is direct bonding by optical contacting and by chemical
activation. Both methods do not require elevated temperatures but highly polished surfaces [244, 245].
Applications requiring pressures higher than atmospheric conditions [246] can be made using steel
chambers in combination with metal flanges joined by active soldering [247].

7.5. Filling of alkali vapor cells

The filling of sealed vapor cells with highly reactive alkalis can be achieved in different manners. The most
common method relies on pure alkali samples, which are transferred from a sealed glass ampule into the cell.
To do so, one or more cells to be filled are connected by glassblowing to a breakseal ampule via a manifold as
shown in figure 17 (some good descriptions and illustrations are also found in [219, 243, 248]). This
manifold is connected to a turbo pump using e.g. a commercially sourced glass flange (we use ISO-KF) that
is joined to the manifold. The glass flange is connected to a metal flange using a standard FKM fluorine
rubber centering o-ring and a clamp made of a material that will not break the glass flange under mechanical
and thermal stress. We use clamps made from FR-2 synthetic resin bonded paper (commercially sourced
from the Pertinax brand material) for this purpose. The material of the o-ring will not be in contact with
large quantities of alkali vapor under vacuum conditions but limits the ultimate leakage performance of the
system compared to metal CF flanges. On the pump side, to provide a suitable amount of flexibility and play

27
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

Figure 17. Photograph of a fully prepared manifold for the filling of an alkali vapor cell. While all elements are to scale, we would
usually leave more space between the flange, the vapor cell, and the breakseal ampule for easier working access and thermal
separation. A 7.5 cm long, empty vapor cell (f) is attached to a glass manifold (inner diameter 16 mm, outer diameter 20 mm)
made of borosilicate glass via a tube (inner diameter 3 mm, outer diameter 6 mm). This manifold ends in an ISO-KF16 flange,
which can be connected to a standard metal ISO-KF16 flange on a vacuum corrugated hose (c) using a centering o-ring (d) and a
Pertinax clamp (a). A few clean steel balls (e) are placed in the manifold and held in place using a strong bar magnet (b). The right
inset shows the breakseal (g) inside a tube (inner diameter 9 mm, outer diameter 6 mm) with a steel ball (e) and the bar magnet
used to break the seal under vacuum. In the lower part of the breakseal, rubidium droplets and the silvery shining bulk material
are visible.

for a large variety of oddly shaped manifolds, we attach the manifolds to a flexible ISO-KF corrugated hose.
The whole assembly and especially the parts that will be in contact with the alkali should then be baked for a
couple of days above 100 ◦ C. A higher bake-out temperature will increase the bake-out speed and remove
additional residual contaminants. The pressure after the bake-out at room temperature should be on the
order of 10 × 10−7 mbar or better. To open the ampule, we use one or multiple small steel balls which get
slammed into the glass nozzle of the breakseal of the ampule with the help of a strong magnet from the
outside. Multiple steel balls can increase the momentum to break particularly sturdy glass nozzles but
increase the danger of accidentally destroying the whole manifold, which is why we often start with a single
steel ball. We typically also place a piece of cloth between the metallic magnet and the glass manifold to
prevent scratching and structurally damaging the glass. The alkali metal is then driven into the cell by gently
heating the metal with a heat gun and shaking the liquid alkali metal into the cell. Finally, the cell is pinched
off while the assembly is still attached to the vacuum pump. In our case this involves two people: One person
holding the manifold assembly steady such that the pinch off location is accessible while mechanically
stabilizing the flange connection, and a scientific glassblower performing the pinch off.
This basic process leads to a macroscopic droplet of the alkali deposited in the vapor cell, which can be
undesirable due to the application, the price of e.g. isotopically enriched rubidium, or the purity of the
available material. It is instead possible to distill the alkali from the breakseal ampule to the vapor cell by
heating the manifold and creating a condensation spot in the vapor cell. In this case we close the connection
to the vacuum pump during distillation e.g. with a metal ball [248] to prevent the loss of the material. We
perform distillation of rubidium at slightly above 100 ◦ C (having the cold spot at room temperature) which
allows us to control the transferred amount by observing the formation of a thin-film rubidium layer over
the course of several minutes to hours. Distillation of cesium is possible even at room temperature by
introducing a cold spot with ice water.
Any buffer gases can be introduced into the vapor cell at this stage by carefully opening a flow from a gas
bottle or gas mixing unit. In this case, a two-step sealing process might be required to ensure predictable
pressure conditions [248]: First, the alkali is brought only close to the final location and then removed from

28
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

the manifold together with the cell as described above. Then, in the already enclosed system, the alkali is
moved or distilled into the actual vapor cell. Such a two-step process is also advantageous if the manifold is
impracticably large or fragile.
For small scale cells, as used for magentometry and atom clocks, this filling method is not very well
controlled and many methods have been reported [218, 219]. Direct transfer or pipetting of molten alkalis
requires an artificial, protective, non-reacting atmosphere. Liew et al [234] used a chemical reaction between
barium azide and alkali chloride which produces at elevated temperature elemental alkali and nitrogen.
Similarly, one can decompose RbN3 with the help of UV light [249]. With both methods it is possible to
adjust a specific partial pressure of nitrogen.
For some applications it is necessary to avoid any buffer gas to reduce collisional broadening and excited
state quenching. In this case one can use small samples of alkalis sealed in a wax layer to avoid oxidation
[250, 251].

Data availability statement

All data that support the findings of this study are included within the article (and any supplementary files).

Acknowledgments

The authors thank Frank Schreiber for the preparation of the glass vapor cells and detailed descriptions and
discussions of the fabrication process. We also thank Patrick Kaspar and Moritz Seltenreich for support
during the setup phase, and Philipp Neufeld for contributions to the code base.
This letter is supported by the Deutsche Forschungsgemeinschaft (DFG) via Grant LO 1657/7-1 under
DFG SPP 1929 GiRyd as well as Grant DFG 431314977/GRK2642 (Research Training Group: ‘Towards
Graduate Experts in Photonic Quantum Technologies’). This project has also received funding from the
European Union’s Horizon 2020 research and innovation program under Grant Agreement No. 820393
(macQsimal).

ORCID iDs

Max Mäusezahl  [Link]

Fabian Munkes  [Link]

References
[1] Knappe S, Shah V, Schwindt P D D, Hollberg L, Kitching J, Liew L-A and Moreland J 2004 Appl. Phys. Lett. 85 1460–2
[2] Matei D G et al 2017 Phys. Rev. Lett. 118 263202
[3] Groswasser D, Waxman A, Givon M, Aviv G, Japha Y, Keil M and Folman R 2009 Rev. Sci. Instrum. 80 093103
[4] Lee C, Iwata G Z, Corsini E, Higbie J M, Knappe S, Ledbetter M P and Budker D 2011 Rev. Sci. Instrum. 82 043107
[5] Dinkelaker A N et al 2017 Appl. Opt. 56 1388
[6] Maurice V, Newman Z L, Dickerson S, Rivers M, Hsiao J, Greene P, Mescher M, Kitching J, Hummon M T and Johnson C 2020
Opt. Express 28 24708–20
[7] Lee S, Lee S-B, Park S E, Hong H-G, Heo M-S, Seo S, Jeong J, Kwon T Y and Moon G 2021 Opt. Lasers Eng. 146 106698
[8] Salomon C, Hils D and Hall J L 1988 J. Opt. Soc. Am. B 5 1576
[9] Alnis J, Matveev A, Kolachevsky N, Udem T and Hänsch T W 2008 Phys. Rev. A 77 053809
[10] Kessler T, Hagemann C, Grebing C, Legero T, Sterr U, Riehle F, Martin M J, Chen L and Ye J 2012 Nat. Photon. 6 687–92
[11] Torrance J S, Sparkes B M, Turner L D and Scholten R E 2016 Opt. Express 24 11396–406
[12] Lee W-K, Park C Y, Heo M-S, Yu D-H and Kim H 2020 Appl. Opt. 59 8918–24
[13] Li Q-X et al 2020 Opt. Lasers Eng. 126 105881
[14] Koch G J 2003 Opt. Eng. 42 1690
[15] Allard F, Maksimovic I, Abgrall M and Laurent P 2004 Rev. Sci. Instrum. 75 54–58
[16] Canuto E, Musso F and Massotti L 2007 IEEE Trans. Ind. Electron. 54 848–57
[17] Sparkes B M, Chrzanowski H M, Parrain D P, Buchler B C, Lam P K and Symul T 2011 Rev. Sci. Instrum. 82 075113
[18] Barwood G P, Gill P, Huang G and Klein H A 2012 Meas. Sci. Technol. 23 055201
[19] Roy A, Sharma L, Chakraborty I, Panja S, Ojha V N and De S 2019 J. Instrum. 14 P05012
[20] Guo X, Zhang L, Liu J, Chen L, Fan L, Xu G, Liu T, Dong R and Zhang S 2022 Opt. Laser Technol. 145 107498
[21] Yang G, Barry J F, Shuman E S, Steinecker M H and DeMille D 2012 J. Instrum. 7 P10026
[22] Huang K, Le Jeannic H, Ruaudel J, Morin O and Laurat J 2014 Rev. Sci. Instrum. 85 123112
[23] Leibrandt D R and Heidecker J 2015 Rev. Sci. Instrum. 86 123115
[24] Jørgensen N B, Birkmose D, Trelborg K, Wacker L, Winter N, Hilliard A J, Bason M G and Arlt J J 2016 Rev. Sci. Instrum. 87 073106
[25] Subhankar S, Restelli A, Wang Y, Rolston S L and Porto J V 2019 Rev. Sci. Instrum. 90 043115
[26] Preuschoff T, Schlosser M and Birkl G 2020 Rev. Sci. Instrum. 91 083001
[27] Strangfeld A, Wiegand B, Kluge J, Schoch M and Krutzik M 2022 Opt. Express 30 12039
[28] Neuhaus L et al 2023 Python Red Pitaya Lockbox (PyRPL): An open source software package for digital feedback control in
quantum optics experiments Re. Sci. Instrum. 95 033003

29
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

[29] White A 1965 IEEE J. Quantum Electron. 1 349–57

[30] Hall J 1968 IEEE J. Quantum Electron. 4 638–41
[31] Baird K M and Hanes G R 1974 Rep. Prog. Phys. 37 927–50
[32] Hall J L 1978 Science 202 147–56
[33] Ohtsu M 1988 Opt. Quantum Electron. 20 283–300
[34] Hamilton M W 1989 Contemp. Phys. 30 21–33
[35] Roberts M, Taylor P and Gill P Laser linewidth at the sub-hertz level NPL Report (Centre for Length Metrology)
[36] Hall J L, Taubman M S and Ye J 2010 Laser stabilization Handbook of Optics vol 2, ed M Bass, V N Mahajan and E van Stryland
(McGraw-Hill) pp 22.1–22.7
[37] Taubman M S 2013 Optical frequency stabilization and optical phase locked loops: golden threads of precision measurement 2013
American Control Conf. (IEEE) pp 1488–505
[38] Wu Y, Sun B and Li X 2021 J. Korean Phys. Soc. 79 795–809
[39] Fang Z, Cai H, Chen G and Qu R 2017 Single Frequency Semiconductor Lasers vol 9 (Springer)
[40] Fox R W, Oates C W and Hollberg L W 2003 Exp. Methods Phys. Sci. 40 1–46
[41] Gawlik W and Zachorowski J 2004 Opt. Appl. 34 607–18
[42] Sullivan D B, Allan D W, Howe D A and Walls F L 1990 NIST Tech. Notes 1337
[43] Zhu M and Hall J L 1993 J. Opt. Soc. Am. B 10 802
[44] Riley W J and Howe D A 2008 Handbook of Frequency Stability Analysis vol 1065 (NIST Special Publicatios)
[45] Allan D W 1966 Proc. IEEE 54 221–30
[46] Bechhoefer J 2005 Rev. Mod. Phys. 77 783–836
[47] Talvitie H, Pietiläinen A, Ludvigsen H and Ikonen E 1997 Rev. Sci. Instrum. 68 1–7
[48] Stockton J, Armen M and Mabuchi H 2002 J. Opt. Soc. Am. B 19 3019
[49] Yu S J, Fajeau E, Liu L Q, Jones D J and Madison K W 2018 Rev. Sci. Instrum. 89 025107
[50] Carminati M and Scandurra G 2021 Rev. Sci. Instrum. 92 091501
[51] Ziegler J G and Nichols N B 1993 J. Dyn. Syst. Meas. Control 115 220–2
[52] Saliba S D and Scholten R E 2009 Appl. Opt. 48 6961–6
[53] Ricci L, Weidemüller M, Esslinger T, Hemmerich A, Zimmermann C, Vuletic V, König W and Hänsch T W 1995 Opt. Commun.
117 541–9
[54] Fleming M and Mooradian A 1981 IEEE J. Quantum Electron. 17 44–59
[55] Bennetts S, McDonald G D, Hardman K S, Debs J E, Kuhn C C N, Close J D and Robins N P 2014 Opt. Express 22 10642–54
[56] Shi W, Fang Q, Zhu X, Norwood R A and Peyghambarian N 2014 Appl. Opt. 53 6554–68
[57] Hänsch T W 1973 Spectroscopy with tunable lasers Atomic Physics 3 ed S J Smith and G K Walters (Springer) pp 579–614
[58] Duarte F J and Hillman L W (eds) 1990 Dye Laser Principles: With Applications (Quantum Electronics – Principles and
Applications) (Academic, Inc.) ([Link]
[59] Walther H 1974 Phys. Scr. 9 297–305
[60] Hollberg L 1990 cw dye lasers Dye Laser Principles (Quantum Electronics – Principles and Applications) ed F J Duarte and
L W Hillman (Academic, Inc.) ([Link]
[61] Wall K F and Sanchez A 1990 Linc. Lab. J. 3 447–62
[62] Duarte F J 2016 Tunable Laser Applications (Optical Science and Engineering) 3rd edn (CRC Press) ([Link]
10.1201/b19508)
[63] Guina M, Rantamäki A and Härkönen A 2017 J. Phys. D: Appl. Phys. 50 383001
[64] Burd S C et al 2023 J. Opt. Soc. Am. B 40 773
[65] Ma L-S, Jungner P, Ye J and Hall J L 1994 Opt. Lett. 19 1777–9
[66] Hall J L, Layer H P and Deslattes R D 1977 IEEE J. Quantum Electron. 13 857–8
[67] Roder H 1931 Proc. IRE 19 2145–76
[68] Harris S E, Oshman M K, McMurtry B J and Ammann E O 1965 Appl. Phys. Lett. 7 185–7
[69] Black E D 2001 Am. J. Phys. 69 79–87
[70] Olver F W J, Daniel L W, Ronald B F and Charles C W 2010 NIST Handbook of Mathematical Functions (Cambridge University
Press and NIST) ([Link]
[71] Marti G E, Hutson R B, Goban A, Campbell S L, Poli N and Ye J 2018 Phys. Rev. Lett. 120 103201
[72] Harvey M D and White A G 2003 Opt. Commun. 221 163–71
[73] Siegman A E 1986 Lasers (University Science Books)
[74] Kogelnik H and Li T 1966 Appl. Opt. 5 1550–67
[75] Chunosov N 2022 Rezonator2 (available at: [Link]
[76] Izumi K, Sigg D and Barsotti L 2014 Opt. Lett. 39 5285–8
[77] Numata K, Kemery A and Camp J 2004 Phys. Rev. Lett. 93 250602
[78] Millo J, Magalhães D V, Mandache C, Le Coq Y, English E M L, Westergaard P G, Lodewyck J, Bize S, Lemonde P and Santarelli G
2009 Phys. Rev. A 79 053829
[79] Boyd J A and Lahaye T 2024 Am. J. Phys. 92 50–58
[80] Barger R L, Sorem M S and Hall J L 1973 Appl. Phys. Lett. 22 573–5
[81] MacAdam K B, Steinbach A and Wieman C 1992 Am. J. Phys. 60 1098–111
[82] Werle P, Mcke R and Slemr F 1993 Appl. Phys. B 57 131–9
[83] Supplee J M, Whittaker E A and Lenth W 1994 Appl. Opt. 33 6294–302
[84] Rowley W R C and Wilson D C 1963 Nature 200 745–7
[85] Taubman M S and Hall J L 2000 Opt. Lett. 25 311–3
[86] Talvitie H, Merimaa M and Ikonen E 1998 Opt. Commun. 152 182–8
[87] Oxley P and Collins P 2010 Appl. Phys. B 101 23–31
[88] Liu X and Boudot R 2012 IEEE Trans. Instrum. Meas. 61 2852–5
[89] Pound R V 1946 Rev. Sci. Instrum. 17 490–505
[90] Drever R W P, Hall J L, Kowalski F V, Hough J, Ford G M, Munley A J and Ward H 1983 Appl. Phys. B 31 97–105
[91] Zhang W et al 2014 Opt. Lett. 39 1980–3
[92] Hansch T W and Couillaud B 1980 Opt. Commun. 35 441–4
[93] Bateman J E, Murray R L D, Himsworth M, Ohadi H, Xuereb A and Freegarde T 2010 J. Opt. Soc. Am. B 27 1530

30
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

[94] Vainio M, Bernard J E and Marmet L 2011 Appl. Phys. B 104 897–908
[95] Hannig S, Mielke J, Fenske J A, Misera M, Beev N, Ospelkaus C and Schmidt P O 2018 Rev. Sci. Instrum. 89 013106
[96] Ludlow A D, Boyd M M, Ye J, Peik E and Schmidt P O 2015 Rev. Mod. Phys. 87 637–701
[97] Demtröder W 2008 Laser Spectroscopy 1 4th edn (Springer) ([Link]
[98] Demtröder W 2015 Laser Spectroscopy 2 5th edn (Springer) ([Link]
[99] Sakai Y, Yokohama I, Kano G and Sudo S 1992 IEEE Photonics Technol. Lett. 4 96–98
[100] Lee P H and Skolnick M L 1967 Appl. Phys. Lett. 10 303–5
[101] Haroche S and Hartmann F 1972 Phys. Rev. A 6 1280–300
[102] Preston D W 1996 Am. J. Phys. 64 1432–6
[103] Schmidt O, Knaak K M, Wynands R and Meschede D 1994 Appl. Phys. B 59 167–78
[104] Ohta Y, Maehara S, Hasebe K, Kurosaki Y, Sato T, Ohkawa M, Maruyama T and Tsubokawa T 2006 Frequency stabilization of a
semiconductor laser using the Rb saturated absorption spectroscopy Proc. SPIE 6115 61150
[105] Wang Q, Duan J, Qi X-H, Zhang Y and Chen X-Z 2015 Chin. Phys. Lett. 32 054206
[106] Genov G, Lellinger T E, Halfmann T and Peters T 2017 J. Opt. Soc. Am. B 34 2018
[107] Wieman C and Hänsch T W 1976 Phys. Rev. Lett. 36 1170–3
[108] Pearman C P, Adams C S, Cox S G, Griffin P F, Smith D A and Hughes I G 2002 J. Phys. B: At. Mol. Opt. Phys. 35 5141–51
[109] Harris M L, Adams C S, Cornish S L, McLeod I C, Tarleton E and Hughes I G 2006 Phys. Rev. A 73 062509
[110] Do H D, Moon G and Noh H-R 2008 Phys. Rev. A 77 032513
[111] Yoshikawa Y, Umeki T, Mukae T, Torii Y and Kuga T 2003 Appl. Opt. 42 6645–9
[112] Ratnapala A, Vale C J, White A G, Harvey M D, Heckenberg N R and Rubinsztein-Dunlop H 2004 Opt. Lett. 29 2704–6
[113] Tiwari V B, Singh S, Mishra S R, Rawat H S and Mehendale S C 2006 Opt. Commun. 263 249–55
[114] Kunz P D, Heavner T P and Jefferts S R 2013 Appl. Opt. 52 8048–53
[115] Lee M W, Jarratt M C, Marciniak C and Biercuk M J 2014 Opt. Express 22 7210–21
[116] Wang Y, Li Y, Wu J, Liu W, Li P, Fu Y, Ma J, Xiao L and Jia S 2022 Front. Phys. 17 22505
[117] Weis A and Derler S 1988 Appl. Opt. 27 2662–5
[118] Chéron B, Gilles H, Hamel J, Moreau O and Sorel H 1994 J. Physique III 4 401–6
[119] Yashchuk V V, Budker D and Davis J R 2000 Rev. Sci. Instrum. 71 341–6
[120] Corwin K L, Lu Z-T, Hand C F, Epstein R J and Wieman C E 1998 Appl. Opt. 37 3295–8
[121] Wasik G, Gawlik W, Zachorowski J and Zawadzki W 2002 Appl. Phys. B 75 613–9
[122] Petelski T, Fattori M, Lamporesi G, Stuhler J and Tino G M 2003 Eur. Phys. J. D 22 279–83
[123] Millett-Sikking A, Hughes I G, Tierney P and Cornish S L 2007 J. Phys. B: At. Mol. Opt. Phys. 40 187–98
[124] Reeves J M, Garcia O and Sackett C A 2006 Appl. Opt. 45 372–6
[125] McCarron D J, Hughes I G, Tierney P and Cornish S L 2007 Rev. Sci. Instrum. 78 093106
[126] Harris M L, Cornish S L, Tripathi A and Hughes I G 2008 J. Phys. B: At. Mol. Opt. Phys. 41 085401
[127] Su D-Q, Meng T-F, Ji Z-H, Yuan J-P, Zhao Y-T, Xiao L-T and Jia S-T 2014 Appl. Opt. 53 7011–6
[128] Pustelny S, Schultze V, Scholtes T and Budker D 2016 Rev. Sci. Instrum. 87 063107
[129] Bjorklund G C 1980 Opt. Lett. 5 15
[130] Schenzle A, DeVoe R G and Brewer R G 1982 Phys. Rev. A 25 2606–21
[131] Bjorklund G C, Levenson M D, Lenth W and Ortiz C 1983 Appl. Phys. B 32 145–52
[132] Cornish S L, Liu Y W, Lane I C, Baird P E G, Barwood G P, Taylor P and Rowley W R C 2000 J. Opt. Soc. Am. B 17 6–10
[133] So C, Spong N L R, Möhl C, Jiao Y, Ilieva T and Adams C S 2019 Opt. Lett. 44 5374–7
[134] Zi F, Wu X, Zhong W, Parker R H, Yu C, Budker S, Lu X and Müller H 2017 Appl. Opt. 56 2649–52
[135] Bell S C, Heywood D M, White J D, Close J D and Scholten R E 2007 Appl. Phys. Lett. 90 171120
[136] Norcia M A and Thompson J K 2016 Rev. Sci. Instrum. 87 023110
[137] Spindeldreier C, Bartosch W, Wendrich T, Rasel E M, Ertmer W and Blume H 2018 FPGA based laser frequency stabilization
using FM spectroscopy Proc. SPIE 10518 1051813
[138] Snyder J J, Raj R K, Bloch D and Ducloy M 1980 Opt. Lett. 5 163–5
[139] Raj R K, Bloch D, Snyder J J, Camy G and Ducloy M 1980 Phys. Rev. Lett. 44 1251–4
[140] Shirley J H 1982 Opt. Lett. 7 537–9
[141] Ducloy M 1982 Opt. Lett. 7 432–3
[142] Eickhoff M L and Hall J L 1995 IEEE Trans. Instrum. Meas. 44 155–8
[143] Sun D, Zhou C, Zhou L, Wang J and Zhan M 2016 Opt. Express 24 10649–62
[144] Bertinetto F, Cordiale P, Galzerano G and Bava E 2001 IEEE Trans. Instrum. Meas. 50 490–2
[145] McCarron D J, King S A and Cornish S L 2008 Meas. Sci. Technol. 19 105601
[146] Zhang J, Wei D, Xie C and Peng K 2003 Opt. Express 11 1338–44
[147] Noh H-R, Park S E, Li L Z, Park J-D and Cho C-H 2011 Opt. Express 19 23444–52
[148] Jaatinen E, Hopper D J and Back J 2009 Meas. Sci. Technol. 20 025302
[149] Riedle E, Ashworth S H, Farrell J T and Nesbitt D J 1994 Rev. Sci. Instrum. 65 42–48
[150] Uetake S, Matsubara K, Ito H, Hayasaka K and Hosokawa M 2009 Appl. Phys. B 97 413–9
[151] Bohlouli-Zanjani P, Afrousheh K and Martin J D D 2006 Rev. Sci. Instrum. 77 093105
[152] Biesheuvel J, Noom D W E, Salumbides E J, Sheridan K T, Ubachs W and Koelemeij J C J 2013 Opt. Express 21 14008–16
[153] Lindsay B G, Smith K A and Dunning F B 1991 Rev. Sci. Instrum. 62 1656–7
[154] Zhao W Z, Simsarian J E, Orozco L A and Sprouse G D 1998 Rev. Sci. Instrum. 69 3737–40
[155] Rossi A, Biancalana V, Mai B and Tomassetti L 2002 Rev. Sci. Instrum. 73 2544–8
[156] Tonyushkin A A, Light A D and Di Rosa M D 2007 Rev. Sci. Instrum. 78 123103
[157] Burke J H T, Garcia O, Hughes K J, Livedalen B and Sackett C A 2005 Rev. Sci. Instrum. 76 116105
[158] Seymour-Smith N, Blythe P, Keller M and Lange W 2010 Rev. Sci. Instrum. 81 075109
[159] Enloe L H and Rodda J L 1965 Proc. IEEE 53 165–6
[160] Preveddelli M, Freegarde T and Hänsch T W 1995 Appl. Phys. B 60 241–8
[161] Stace T, Luiten A N and Kovacich R P 1998 Meas. Sci. Technol. 9 1635–7
[162] Schünemann U, Engler H, Grimm R, Weidemüller M and Zielonkowski M 1999 Rev. Sci. Instrum. 70 242–3
[163] McFerran J J 2018 Opt. Lett. 43 1475–8
[164] Reynolds F C and McFerran J J 2019 Appl. Opt. 58 3128–32

31
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

[165] Strauß N, Ernsting I, Schiller S, Wicht A, Huke P and Rinkleff R-H 2007 Appl. Phys. B 88 21–28
[166] Ritt G, Cennini G, Geckeler C and Weitz M 2004 Appl. Phys. B 79 363–5
[167] Schilt S, Matthey R, Kauffmann-Werner D, Affolderbach C, Mileti G and Thévenaz L 2008 Appl. Opt. 47 4336–44
[168] Uehara T, Tsuji K, Hagiwara K and Onodera N 2014 Opt. Eng. 53 124109
[169] Lipka M, Parniak M and Wasilewski W 2017 Appl. Phys. B 123 238
[170] Li V, Diorico F and Hosten O 2022 Phys. Rev. Appl. 17 054031
[171] Höckel D, Scholz M and Benson O 2009 Appl. Phys. B 94 429–35
[172] Ivanov E N, Esnault F-X and Donley E A 2011 Rev. Sci. Instrum. 82 083110
[173] Kawalec T and Bartoszek-Bober D 2013 Opt. Eng. 52 126105
[174] Cacciapuoti L, de Angelis M, Fattori M, Lamporesi G, Petelski T, Prevedelli M, Stuhler J and Tino G M 2005 Rev. Sci. Instrum.
76 053111
[175] Wang X-L, Tao T-J, Cheng B, Wu B, Xu Y-F, Wang Z-Y and Lin Q 2011 Chin. Phys. Lett. 28 084214
[176] Xu Z, Zhang X, Huang K and Lu X 2012 Rev. Sci. Instrum. 83 093104
[177] Hänsch T W 2006 Rev. Mod. Phys. 78 1297–309
[178] Cundiff S T and Ye J 2003 Rev. Mod. Phys. 75 325–42
[179] Diddams S A 2010 J. Opt. Soc. Am. B 27 B51
[180] Fortier T and Baumann E 2019 Commun. Phys. 2 153
[181] Jost J, Hall J and Ye J 2002 Opt. Express 10 515–20
[182] Schibli T R, Minoshima K, Hong F-L, Inaba H, Bitou Y, Onae A and Matsumoto H 2005 Opt. Lett. 30 2323–5
[183] Park S E, Kim E B, Park Y-H, Yee D S, Kwon T Y, Park C Y, Moon H S and Yoon T H 2006 Opt. Lett. 31 3594–6
[184] Inaba H, Ikegami T, Hong F-L, Bitou Y, Onae A, Schibli T R, Minoshima K and Matsumoto H 2006 Appl. Opt. 45 4910–5
[185] Prehn A, Glöckner R, Rempe G and Zeppenfeld M 2017 Rev. Sci. Instrum. 88 033101
[186] Akamatsu D, Nakajima Y, Inaba H, Hosaka K, Yasuda M, Onae A and Hong F-L 2012 Opt. Express 20 16010–6
[187] Argence B, Chanteau B, Lopez O, Nicolodi D, Abgrall M, Chardonnet C, Daussy C, Darquié B, Le Coq Y and Amy-Klein A 2015
Nat. Photon. 9 456–60
[188] Yasuda M, Kohno T, Inaba H, Nakajima Y, Hosaka K, Onae A and Hong F-L 2010 J. Opt. Soc. Am. B 27 1388
[189] Fordell T, Wallin A E, Lindvall T, Vainio M and Merimaa M 2014 Appl. Opt. 53 7476–82
[190] Gunton W, Semczuk M and Madison K W 2015 Opt. Lett. 40 4372–5
[191] Kowalski F V, Hawkins R T and Schawlow A L 1976 J. Opt. Soc. Am. 66 965–6
[192] Fox P J, Scholten R E, Walkiewicz M R and Drullinger R E 1999 Am. J. Phys. 67 624–30
[193] Scholl T J, Rehse S J, Holt R A and Rosner S D 2004 Rev. Sci. Instrum. 75 3318–26
[194] Castell R, Demtrder W, Fischer A, Kullmer R, Weickenmeier H and Wickert K 1985 Appl. Phys. B 38 1–10
[195] Gardner J L 1985 Appl. Opt. 24 3570
[196] Gray D F, Smith K A and Dunning F B 1986 Appl. Opt. 25 1339
[197] Verlinde M et al 2020 Appl. Phys. B 126 85
[198] König K, Imgram P, Krämer J, Maaß B, Mohr K, Ratajczyk T, Sommer F and Nörtershäuser W 2020 Appl. Phys. B 126 86
[199] Kobtsev S, Kandrushin S and Potekhin A 2007 Appl. Opt. 46 5840–3
[200] Couturier L, Nosske I, Hu F, Tan C, Qiao C, Jiang Y H, Chen P and Weidemüller M 2018 Rev. Sci. Instrum. 89 043103
[201] Saakyan S A, Sautenkov V A, Vilshanskaya E V, Vasiliev V V, Zelener B B and Zelener B V 2015 Quantum Electron. 45 828–32
[202] Ghadimi M, Bridge E M, Scarabel J, Connell S, Shimizu K, Streed E and Lobino M 2020 Appl. Opt. 59 5136–41
[203] Kim J, Kim K, Lee D, Shin Y, Kang S, Kim J R, Choi Y, An K and Lee M 2021 Sensors 21 6255
[204] Thorpe J I, Numata K and Livas J 2008 Opt. Express 16 15980–90
[205] Cuneo C J, Maki J J and McIntyre D H 1994 Appl. Phys. Lett. 64 2625–7
[206] Buczek C J, Freiberg R J and Skolnick M L 1973 Proc. IEEE 61 1411–31
[207] Lang R 1982 IEEE J. Quantum Electron. 18 976–83
[208] Pultinevicius E, Rockenhäuser M, Kogel F, Groß P, Garg T, Prochnow O E and Langen T 2023 Rev. Sci. Instrum. 94 103004
[209] Tomschitz C 2018 A photoionization scheme to create cold ionic impurities from Rydberg atoms MSc Thesis University of
Stuttgart
[210] Neuhaus L, Metzdorff R, Chua S, Jacqmin T, Briant T, Heidmann A, Cohadon P-F and Deléglise S 2017 PyRPL (Python Red
Pitaya Lockbox) FPGA-controlled quantum optics experiments 2017 Conf. on Lasers and Electro-Optics Europe & European
Quantum Electronics Conf. (CLEO/Europe-EQEC) (IEEE) ([Link]
[211] Okoshi T, Kikuchi K and Nakayama A 1980 Electron. Lett. 16 630
[212] Ludvigsen H, Tossavainen M and Kaivola M 1998 Opt. Commun. 155 180–6
[213] Huang S, Zhu T, Liu M and Huang W 2017 Sci. Rep. 7 41988
[214] Gray J E and Allan D W 1974 A method for estimating the frequency stability of an individual oscillator 28th Annual Symp. on
Frequency Control pp 243–6
[215] Premoli A and Tavella P 1993 IEEE Trans. Instrum. Meas. 42 7–13
[216] Compton A H 1931 Electric lamp US Patent US1830312
[217] Compton A H 1926 Glass and article made therefrom US Patent US1570876
[218] Karlen S 2017 Fabrication and characterization of MEMS alkali vapor cells used in chip-scale atomic clocks and other atomic
devices PhD Thesis Université de Neuchâtel
[219] Knapkiewicz P 2018 Micromachines 10 25
[220] de Jong B H W S, Beerkens R G C and van Nijnatten P A 2003 Glass Ullmann’s Encyclopedia of Industrial Chemistry (Wiley)
([Link]
[221] Sarkisyan D, Bloch D, Papoyan A and Ducloy M 2001 Opt. Commun. 200 201–8
[222] Peyrot T, Sortais Y R P, Browaeys A, Sargsyan A, Sarkisyan D, Keaveney J, Hughes I G and Adams C S 2018 Phys. Rev. Lett.
120 243401
[223] Slabinski V J and Smith R L 1971 Rev. Sci. Instrum. 42 1334–8
[224] Vidal C R and Cooper J 1969 J. Appl. Phys. 40 3370–4
[225] Franzen W 1959 Phys. Rev. 115 850–6
[226] Chi H, Quan W, Zhang J, Zhao L and Fang J 2020 Appl. Surf. Sci. 501 143897
[227] Woetzel S, Talkenberg F, Scholtes T, IJsselsteijn R, Schultze V and Meyer H-G 2013 Surf. Coat. Technol. 221 158–62
[228] Perez M A, Nguyen U, Knappe S, Donley E A, Kitching J and Shkel A M 2009 Sensors Actuators A 154 295–303

32
New J. Phys. 26 (2024) 105002 M Mäusezahl et al

[229] Giorgi T A, Ferrario B and Storey B 1985 J. Vac. Sci. Technol. A 3 417–23
[230] Brandt S, Nagel A, Wynands R and Meschede D 1997 Phys. Rev. A 56 R1063–6
[231] Moore J H, Davis C C, Coplan M A and Greer S C 2009 Building Scientific Apparatus 4th edn (Cambridge University Press)
([Link]
[232] Ramm P, Lu J J Q and Taklo M M V () 2012 Handbook of Wafer Bonding (Wiley-VCH) ([Link]
[233] Wallis G and Pomerantz D I 1969 J. Appl. Phys. 40 3946–9
[234] Liew L-A, Knappe S, Moreland J, Robinson H, Hollberg L and Kitching J 2004 Appl. Phys. Lett. 84 2694–6
[235] Daschner R, Ritter R, Kübler H, Frühauf N, Kurz E, Löw R and Pfau T 2012 Opt. Lett. 37 2271–3
[236] Barredo D, Kübler H, Daschner R, Löw R and Pfau T 2013 Phys. Rev. Lett. 110 123002
[237] Sarkisyan D G and Melkonyan A V 1989 Instrum. Exp. Tech. 32 485–6
[238] McVey C I 1979 J. Illum. Eng. Soc. 8 72–77
[239] Gavrilov S A, Gavrish S V and Puchnina S V 2019 Glass Ceram. 75 408–12
[240] Karlen S, Haesler J, Overstolz T, Bergonzi G and Lecomte S 2020 J. Microelectromech. Syst. 29 95–99
[241] Knappe S, Velichansky V, Robinson H G, Kitching J and Hollberg L 2003 Rev. Sci. Instrum. 74 3142–5
[242] Losev S S, Sevostianov D I, Vassiliev V V and Velishansky V L 2015 Phys. Proc. 71 242–6
[243] Maurice V et al 2022 Microsyst. Nanoeng. 8 129
[244] Peyrot T, Beurthe C, Coumar S, Roulliay M, Perronet K, Bonnay P, Adams C S, Browaeys A and Sortais Y R P 2019 Opt. Lett.
44 1940–3
[245] Cutler T F, Hamlyn W J, Renger J, Whittaker K A, Pizzey D, Hughes I G, Sandoghdar V and Adams C S 2020 Phys. Rev. Appl.
14 034054
[246] Larson B, Häusser O, Delheij P P, Whittal D M and Thiessen D 1991 Phys. Rev. A 44 3108–18
[247] Ockenfels T, Vewinger F and Weitz M 2021 Rev. Sci. Instrum. 92 065109
[248] Missout G and Vanier J 1975 IEEE Trans. Instrum. Meas. 24 180–4
[249] Karlen S, Gobet J, Overstolz T, Haesler J and Lecomte S 2017 Opt. Express 25 2187–94
[250] Radhakrishnan S and Lai A 2005 Alkali metal-wax micropackets for chip-scale atomic clocks 13th Int. Conf. on Solid-State Sensors,
Actuators and Microsystems, 2005. Digest of Technical Papers. TRANSDUCERS ’05 (IEEE) pp 23–26
[251] Wang X, Ye M, Lu F, Mao Y, Tian H and Li J 2022 Biosensors 12 165

33