PROCEEDINGS OF SPIE

[Link]/conference-proceedings-of-spie

Comparison of freeform
manufacturing techniques in the
production of monolithic lens arrays

Gregg Davis, Jeffrey Roblee, Alan Hedges

Gregg E. Davis, Jeffrey W. Roblee, Alan R. Hedges, "Comparison of freeform

manufacturing techniques in the production of monolithic lens arrays," Proc.
SPIE 7426, Optical Manufacturing and Testing VIII, 742605 (24 August 2009);
doi: 10.1117/12.824451

Event: SPIE Optical Engineering + Applications, 2009, San Diego, California,

United States

Downloaded From: [Link] on 24 Oct 2023 Terms of Use: [Link]

 Comparison of Freeform Manufacturing Techniques in the Production
of Monolithic Lens Arrays
Gregg E. Davis*a, Jeffrey W. Robleeb, Alan R. Hedgesa
a
II-VI Incorporated, 375 Saxonburg Blvd, Saxonburg, PA, USA 16056
b
Ametek Precitech, 44 Blackbrook Rd, Keene, NH, USA 03431

ABSTRACT

Monolithic lens arrays are used in applications such as hyper-spectral imaging, Shack-Hartmann wavefront sensors, and
lens replication molds, where lens-to-lens registration is critical. Traditionally, monolithic lens arrays are produced by
diamond turning one lens at a time on axis. This process requires the substrate to be shifted to a new position before the
next lens is machined. This intermediate step increases production time and makes it difficult to achieve lens-to-lens
registration accuracy. Freeform diamond machining allows lens arrays to be produced in a single setup. Since there are
no intermediate shifts of the substrate, the lens-to-lens registration is inherent to the program and machine accuracy. The
purpose of this paper is to compare different freeform manufacturing processes in the production of a three-element
germanium lens array. Freeform machining technologies including Slow Tool Servo (STS), Fast Tool Servo (FTS) and
Diamond Micro-Milling (DMM) will be used to produce this lens array. The results for process times, figure, and finish
characteristics will be compared across all three techniques.

Keywords: diamond turning, arrays, hyper-spectral, Shack-Hartmann, micro-milling, slow tool servo, fast tool servo

1. INTRODUCTION
Diamond turning shops have several options for producing monolithic lens arrays. They can produce these arrays on-
axis, one lens at a time, or they can choose from three different freeform diamond turning technologies to produce lens
arrays in a single setup. The freeform, single setup option is enticing since it eliminates a number of labor-intensive
production steps and produces lens-to-lens registration under one micron. For this paper, freeform technologies
including Slow Tool Servo, Fast Tool Servo, and Diamond Micro-Milling were used to produce a benchmark three-
element germanium lens array to show the benefits and drawbacks of each technology.
1.1 Slow Tool Servo [Link]
STS is a relatively low-cost upgrade to a standard two-axis diamond turning lathe. A high-resolution C-axis encoder and
some software is all that is needed to convert from two axes (X and Z) to three (X, Z and C). In most STS applications
the Z-axis slides back and forth as X and C maintain constant velocities. It is slow because the Z-axis is massive and the
linear motor that drives it produces a limited amount of force. Programming of the toolpath is done offline using custom
or commercial Computer-Aided-Manufacturing (CAM) software, and the file is loaded onto the machine prior to
operation. The machine uses spline interpolation to bridge the space between program points which improves finish and
allows for acceleration feed-forward in the control loop. The key to efficient use of the STS (maximizing spindle speed,
while minimizing Z-axis acceleration) lies in programming a smooth toolpath and optimizing the control parameters.
Some advanced CAM software such as DIFFSYS have tools to model accelerations which aid in the process
optimization.
1.2 Fast Tool Servo [Link]
FTS is a separate axis (W) that can be added to a standard diamond turning machine. They are typically actuated by
either a piezo electric stack or a voice coil depending on the desired range of motion. With low mass and high force,
these units can have a bandwidth of 1000Hz and achieve peak accelerations of 30g. The programming can be more
complicated since the program must calculate the stroke (W) as a function of X and C in real-time. Tool tip radius
compensation must also be calculated in real time. A recently developed system for generating FTS commands, called
FastCom, greatly eases this programming task, and it functions with DIFFSYS.

Optical Manufacturing and Testing VIII, edited by James H. Burge, Oliver W. Fähnle, Ray Williamson,
Proc. of SPIE Vol. 7426, 742605 · © 2009 SPIE · CCC code: 0277-786X/09/$18 · doi: 10.1117/12.824451

Proc. of SPIE Vol. 7426 742605-1

Downloaded From: [Link] on 24 Oct 2023

Terms of Use: [Link]
 1.3 Diamond Micro-Milling [Link]
DMM replaces the standard diamond tool with a high-speed milling spindle (>50krpm). It is often configured on a five-
axis machine (X, Y, Z, B and C), but only three axes are necessary. For ease of programming, DMM has often utilized
X, Y and Z axes, but we chose an X, Z and C process because of its higher inherent positioning accuracy. Perhaps most
critical to DMM is the diamond tool. The diamond must be centered perfectly on a cylindrical carbide shank, which in
turn, must be perfectly centered in the collet chuck on the milling spindle. Without this alignment, the cutting swath
would be toroidal instead of spherical. Toolpath file size can be kept relatively small by taking advantage of a special
feature on Precitech machine controllers called para-macro programming. This feature allows special computer code,
similar to BASIC, to be included in the toolpath program. Since we are producing arrayed structures, the toolpath for
only one lens is required. By means of a para-macro program, a coordinate transformation is done in real time to shift
this local lens toolpath anywhere on the surface of the workpiece.

2. LENS ARRAY DEFINITION

We defined the lens array to be exemplary of an infrared hyper-spectral application. The design has three evenly-spaced
convex lenses. The radius of curvature is 10mm and the clear aperture of each lens is 5mm, which gives a sag of
0.318mm. We allowed for a 7mm machining aperture in order to smoothly bridge the gaps between lenses. We chose
germanium as a material since it is typical for infrared imaging optics.

Fig. 1. The FTS and STS version of the lens array design is shown on the left. The clear apertures, shown in red, blend into
a flat 0.450mm below the peak of the lenses. The DMM version of the design is shown on the right. The clear
aperture blends into the micro-milling tool radius. The flat surrounding each lens is for illustration purposes and was
not cut by DMM.

3. DIAMOND TURNING PARAMETERS

3.1 STS Diamond Turning Parameters
The STS part was machined at II-VI using a Precitech Nanoform 250 Ultra. We used a 0.050mm radius tool from
Contour Fine Tooling with enough front clearance to avoid undercutting the lens array surface. The effective rake angle
and front clearance angle varied by +/- 15° at the point of contact due to the surface slopes. We were not sure how this
would affect finish so we conservatively set the finish spiral feed at 0.0005mm/rev. The spindle speed was set to a
constant 50rpm after careful optimization of the control parameters. At this speed, during the highest accelerations of
the cut (0.013g), the Z-axis following error did not exceed +/- 1 fringe @ HeNe (+/-0.0003164mm). The peak
acceleration occurs in the blend zone outside the clear aperture. This zone has a local radius of curvature of about 4mm,
which means the Z acceleration here is about 2.5 times larger than the acceleration in the clear aperture. A larger lens
spacing would permit a higher cutting speed. We started each pass 9.3mm from center which left enough room to slope
down to the bottom flat and begin cutting the lenses. Table 1 shows a record of the passes taken to produce this part as
well as the calculation of total processing time, which was 725.4 minutes (~11.5 hours).

Proc. of SPIE Vol. 7426 742605-2

Downloaded From: [Link] on 24 Oct 2023

Terms of Use: [Link]
 Table 1. Record of STS cutting passes.

Cut Depth Spiral Feed Time Total Time

Passes (mm) (mm/rev) RPM (min/pass) (min)
rough 9 0.0508 0.0100 50 18.6 167.4
semi-rough 1 0.0381 0.0020 50 93.0 93.0
semi-finish 1 0.0127 0.0020 50 93.0 93.0
finish 1 0.0025 0.0005 50 372.0 372.0
725.4

3.2 FTS Diamond Turning Parameters

The FTS part was machined at Precitech using a Nanoform 250 Ultra and a Jonix FTS unit with 500μm of stroke. We
used the same diamond tool that we used for the STS part, except we modified shank to fit in the FTS unit. The
diamond had been relapped so there was no tool wear at the beginning. The tool was lightly used on one setup lens array
cut prior to this sample. We maintained the finish spiral feed of 0.0005mm/rev, however we increased the spindle speed
by a factor of ten to 500rpm. At this speed the maximum acceleration was 100 times larger at 1.26g. The real-time
following error of the FTS unit was not recorded as this is not a built-in option on the FTS controller. The FTS portion
of the cut began 9.3mm from center just like the STS part. Table 2 shows a record of the passes taken to produce this
part as well as the calculation of total processing time, which was 53.9 minutes (~0.9 hours). Note the more aggressive
cutting depths used here compared to Table 1.
Table 2. Record of FTS cutting passes.

Cut Depth Spiral Feed Time Total Time

Passes (mm) (mm/rev) RPM (min/pass) (min)
rough 4 0.1000 0.0100 500 1.9 7.4
semi-finish 1 0.0550 0.0020 500 9.3 9.3
finish 1 0.0040 0.0005 500 37.2 37.2
53.9

3.3 DMM Diamond Turning Parameters

The DMM part was machined at II-VI using a Precitech Nanoform 700 Ultra. The machine is configured with X, Z, and
C axes, as well as a rotary B-axis. A 50krpm milling spindle from Professional Instruments was mounted to the B-axis
and aligned 20° in the horizontal plane relative to the work spindle axis (C axis). A custom local spiral milling program
was used to command the machine. This machine does not have a Y-axis so the local spiral path was generated in
cylindrical coordinates (X, C, Z) instead of Cartesian coordinates (X, Y, Z). We used a 0.5mm radius ball end mill from
Chardon Tool. The tool was not in pristine condition as several array samples had been cut previously with it. Table 3
shows a record of the single pass taken to produce this part as well as the calculation of total processing time, which was
169.2 minutes (~2.9 hours).
Table 3. Record of DMM cutting passes.

Cut Depth Spiral Feed Max Linear Time per Lens Total Time
Passes (mm) (mm/rev) Feed (mm/min) (min/lens) (min)
finish 1 <0.450 0.0040 100 56.4 169.2
169.2

Proc. of SPIE Vol. 7426 742605-3

Downloaded From: [Link] on 24 Oct 2023

Terms of Use: [Link]
 4. RESULTS
We measured radius, irregularity and roughness for all three parts. Radius and irregularity for each lens were measured
with a Zygo GPI interferometer, F1.5 transmission sphere and a distance measuring interferometer. Roughness
measurements were taken with a Zygo NewView5000. We used the advanced texture application within MetroPro to
analyze the roughness data. We filtered out wavelengths larger than 40μm as this is more of a form issue for infrared
light under 10μm. Cutting results for all three parts are listed in Table 4.
Table 4. Results for radius, irregularity and roughness. Irregularity is reported using 98% clear aperture (4.9mm).

Radius Irregularity Roughness,

(mm) (fringes @ HeNe) Rq (Å rms)
STS top 9.993 5.05 center 6.3
left 9.993 4.93 edge1 7.8
right 9.993 5.03 edge2 14.8
mean 9.993 5.00 mean 9.6
std 0.000 0.06 std 4.5

Radius Irregularity Roughness,

(mm) (fringes @ HeNe) Rq (Å rms)
FTS top 10.008 6.33 center 39.8
left 10.010 6.35 edge1 170.0
right 10.010 6.34 edge2 32.0
mean 10.009 6.34 mean 80.6
std 0.001 0.01 std 77.5

Radius Irregularity Roughness,

(mm) (fringes @ HeNe) Rq (Å rms)
DMM top 9.981 2.08 center 40.7
left 9.996 1.94 edge1 138.4
right 9.990 2.14 edge2 108.6
mean 9.989 2.05 mean 95.9
std 0.008 0.10 std 50.1

The radius of curvature was repeatable from lens to lens on the STS and FTS parts. This is attributable to each lens
being produced at the same time. The DMM radii were more variable. This is likely due to thermal variations of the
milling spindle during each individual lens cut. The DMM lenses were cut in the sequence listed in Table 4.
The shape of the irregularity, shown in Figs. 2, 4 and 6, was nearly identical from lens to lens with a 120° rotation.
Using a 98% clear aperture (4.9mm) the irregularity was 1.58μm for the STS part, 2.01μm for FTS and 0.65μm for the
DMM. Apart from some localized errors at the edges of the clear apertures of the FTS and STS parts, astigmatism is a
large component of the error for all three parts. This suggests a common error source between these different processes.
Misalignment of the cutting tool with respect to the centerline of the C-axis is the likely source of this error.
The STS process resulted in superior roughness averaging 9.6Å rms. The number of passes and depths of cut were more
conservative than the other methods. The STS process also benefited from a two-tool process, while the FTS and DMM
processes used the same tool for roughing and finishing. The FTS and DMM processes resulted in an average roughness
of 80.6Å and 95.9Å rms respectively. Both STS and FTS processes showed increased roughness on the downhill slope
of each lens. This is where the effective rake angle was most positive, which may be the cause of increased sub-surface
damage (pits) shown in Figs. 3 and 5. Deeper finish cuts could help reduce this damage if it occurred during the semi-
finish cut. By coincidence, the downhill slopes were roughly aligned with the three-spoke pattern that resulted from the
1-1-1 crystal orientation of the germanium. The DMM process, shown in Fig. 7, resulted in surprisingly good roughness
considering the milling tool had a 0° rake and was partially worn. Roughness was as low as 40.7Å rms.

Proc. of SPIE Vol. 7426 742605-4

Downloaded From: [Link] on 24 Oct 2023

Terms of Use: [Link]
 4.1 STS Results

Irr ~ 5.0 fr

Fig. 2. A photo of the STS lens array shows the final product (left). Irregularity of all three lenses shows a repeatable figure
error of ~5.0 fringes @ HeNe (center). An interferogram of the flat shows the overshoot of the Z-axis as the trailing
edge of each lens rotated past (right). The spindle speed was 50 rpm in the counter-clockwise direction.

Fig. 3. The rms roughness (Rq) of the top lens is shown at three locations: center, right and left. The Rq value is calculated
after long wavelength errors (> 40 μm) are filtered from the original data. Since this part was spun counter-clockwise,
the left location was an uphill climb for the tool, which is important when considering the varying relative rake angle of
the tool. The downhill location (at the right edge) shows a slight increase in sub-surface damage. The edge of the
crystal orientation was also aligned with this region.

Proc. of SPIE Vol. 7426 742605-5

Downloaded From: [Link] on 24 Oct 2023

Terms of Use: [Link]
 4.2 FTS Results

Irr ~ 6.3 fr

Fig. 4. A photo of the FTS lens array looks very similar to the STS version shown previously (left). Irregularity of all three
lenses shows a repeatable figure error of ~6.3 fringes @ HeNe (center). An interferogram of the flat shows the
overshoot of the W-axis as the trailing edge of each lens rotated past (right). The spindle speed was 500rpm in the
clockwise direction.

Fig. 5. The rms roughness (Rq) of the top lens is shown at three locations: center, right and left. The left location was a
downhill slope for the tool due to clockwise rotation of this part. The downhill location shows a large increase in
surface damage possibly due to the relative rake angle. Crystal orientation lined up with this region as well.

Proc. of SPIE Vol. 7426 742605-6

Downloaded From: [Link] on 24 Oct 2023

Terms of Use: [Link]
 4.3 DMM Results

Irr ~ 2.1 fr

Fig. 6. A photo of the DMM lens array looks very different from the STS and FTS versions due to the local spiral milling
process (left). The irregularity plots of all three lenses show a repeatable figure error of ~2.1 fringes @ HeNe (center).
An interferogram of the flat, cut in a separate diamond turning step, shows the footprint of the DMM apertures. The
milling spindle was angled 20º in the horizontal plane and run at 50,000rpm.

Fig. 7. The rms roughness (Rq) of the top lens is shown at three locations: center, right and left.

Proc. of SPIE Vol. 7426 742605-7

Downloaded From: [Link] on 24 Oct 2023

Terms of Use: [Link]
 5. CONCLUSIONS
Each process produced the three-element germanium lens array to a level of quality that could be considered functional
for hyper-spectral imaging in the infrared. The processing times were considerably different from each other by an order
of magnitude, though some sacrifice of surface quality was required for each process. The surface figure on all three
processes has the potential to be improved through a freeform correction. This was not attempted for this paper, but in
practice the processing times should remain unaffected after the first iteration. Roughness, however, will be limited by
one-tool processes, surface slopes, available tool geometry and crystal orientation. The faster FTS and DMM single-tool
processes may lose their processing time advantages for large batch sizes if they are constantly delayed by new tool
setups and freeform corrections. Tool life was not a focus of testing for this paper so it remains a question for the FTS
and DMM processes.
The STS process gives superior roughness results, but is roughly ten times slower than the FTS process and three times
slower than the DMM process. The quality of the finish suggests that a larger feed/rev could be used. This could reduce
processing time perhaps by a factor of two while maintaining acceptable finish. Cutting time can be reduced further by
using more aggressive cutting parameters, like those used for the FTS cuts. The irregularity was high on the STS part,
but very repeatable. Implementation of a freeform correction should be feasible to improve this result. Scaling of this
process for large arrays is likely to be time-prohibitive.
The FTS process also yields repeatable but high irregularity. A modest reduction in rpm, or a better optimized tool path
outside the clear aperture, will yield better results while maintaining a clear processing time advantage. The results
shown in Figure 4 support this conclusion. If the regions near the leading and trailing edges of the lens are excluded, the
figure error drops to 1.2μm. Implementation of a freeform correction is likely to be challenging since all of the
calculations must be done in real-time. Improvements in tool radius compensation will also lead to better figure. The
FTS process should scale reasonably well for producing large arrays provided the geometry is within the stroke range.
DMM shows the best irregularity results of all three processes. A freeform correction should be relatively easy to
implement and push irregularity results below one fringe. Processing time was three times faster than the STS process.
Development of a negative rake milling tool is likely to extend tool life and improve finish considerably on germanium.
The DMM process is more versatile, and it scales better than the other processes for producing large, sparse arrays since
it only cuts the lens aperture area.
Clearly more development is required to realize the full potential of these freeform manufacturing technologies. Key
areas of interest include form error correction, diamond tool geometry, tool setup procedures, and process parameters.
Improvements in precision diamond turning machines and metrology equipment will also increase the potential of
freeform manufacturing. Advances in any of these areas will bolster the success of freeform manufacturing and
facilitate growth in the design and application of optical arrays.

Proc. of SPIE Vol. 7426 742605-8

Downloaded From: [Link] on 24 Oct 2023

Terms of Use: [Link]