Advanced microprocessor systems

Advanced Microprocessor System Block Diagram

Up to 1KW, water cooling 1.8 – 3.8 GHz

Power ROM/FLASH
Clock

Dual SLI PCI express DDR1,2,3,4-SDRAM

Graphics CPU   RAM  
L1/L2/L3  

Hypertransport/PCI express

3D sound,
Network Comm.
10/100bT/Gb/802.11
I/O Control A/V Joystick,
USB
Wireless, Ethernet, Optical IEEE488
Bluetooth
SCSI
IDE/ATA/SATA
FDDI
Peripherals
Mass Storage
Keyboard, Mouse, scanner, USB
SATA, RAID, DVD,
Blueray
CPU  Advancement  Mechanisms:  
 
1)  Advances  in  processor  architecture:  
           
 RISC  
           Pipelining  
           Superscalar  
           Out-­‐of-­‐order  execu9on  
 
 
RISC  -­‐  a  system  that  uses  a  small,  highly-­‐opBmized  set  of  instrucBons  that  typicaly  
execute  in  one  clock  cycle,  rather  than  a  more  specialized  set  of  instrucBons  that  may  
require  several  clock  cycles.  There  are  many  traits  associated  with  RISC.  For  example,  
most  machines  are  implemented  with  mulBple  internal  buses  similar  to  a  Harvard  
architecture.  Another  common  trait  is  that  RISC  systems  use  the  load/store  
architecture,  where  memory  is  normally  accessed  only  through  specific  instrucBons,  
rather  than  accessed  as  part  of  other  instrucBons  like  an  add.    
Pipelining  -­‐  a  technique  used  in  the  design  of  computers  to  increase  their  
instrucBon  throughput.  Rather  than  processing  each  instrucBon  sequenBally,  
each  instrucBon  is  split  up  into  a  sequence  of  steps  which  are  executed  within  a  
small  offset  from  one  another.  Thus,  different  steps  can  be  executed  concurrently  
(by  different  circuitry),  and  almost  in  parallel.  
Superscalar  -­‐  a  form  of  parallelism  called  instrucBon-­‐level  parallelism  within  a  single  
processor.  A  CPU  executes  more  than  one  instrucBon  during  a  clock  cycle  by  
simultaneously  dispatching  mulBple  instrucBons  to  redundant  funcBonal  units  on  the  
processor.    
Out-­‐of-­‐order  execu9on  –  A  CPU  technique  involving:      
 
• fetching  instrucBons  in  a  compiler-­‐generated  order  

• The  instrucBons  are  dynamically  scheduled    

 
•  In  between  they  may  be  executed  in  some  other  order  
 
•  Independent  instrucBons  behind  a  stalled  instrucBon  can  pass  it  
 
•  InstrucBons  are  reshuffled  back  into  the  correct  order  for  writeback  stage  
CPU  Advancement  Mechanisms:  
 
2)  Advances  in  processor  manufacturing:  
         Lithography  processes:  
                       
   Op9cal  
                       e-­‐beam  
                       X-­‐ray  
                       Extreme  UV  
 
 
Op9cal  Lithography  -­‐a  photograhic  process  used  to  paVern  parts  of  a  semiconductor  
film  or  the  materialof  a  substrate.  It  uses  light  to  transfer  a  geometric  paVern  from  a  
photomask  to  a  light-­‐sensiBve  chemical  "photoresist”,  on  the  substrate.  A  series  of  
chemical  treatments  then  either  engraves  the  exposure  paVern,  or  enables  deposiBon  
of  a  new  material  in  the  desired  paVern.  
e-­‐beam  Lithography  -­‐  is  the  pracBce  of  scanning  a  focused  beam  of  electrons  to  draw  
custom  shapes  on  a  surface  covered  with  an  electron  sensiBve  film  called  a  resist.  The  
purpose,  as  with  photolithography,  is  to  create  very  small  structures  in  the  resist  that  can  
subsequently  be  transferred  to  the  substrate  material,  oYen  by  etching.  Using  this  
technique,  sub  10nm  structures  can  be  created.  
X-­‐ray  Lithography  -­‐  uses  X-­‐rays  to  transfer  a  geometric  paVern  from  a  mask  to  a  light-­‐
sensiBve  chemical  photoresist  on  a  substrate.  A  series  of  chemical  treatments  then  
engraves  the  produced  paVern  into  the  material  underneath  the  photoresist.  Using  this  
technique,  sub  1nm  structures  can  be  created.  
Extreme  UV  Lithography  -­‐  a  next-­‐generaBon  lithography  technology  using  an  extreme  
ultraviolet  (EUV)  wavelength,  currently  expected  to  be  around  13.5  nm.  Using  this  
technique,  sub  20nm  structures  can  be  fabricated.    
CPU  Advancement  Mechanisms:  
 
3)  Advances  in  processor  implementa2ons:  
           
 Semiconductor  compu9ng  
           Op9cal  compu9ng  
           Molecular  compu9ng  
           Quantum  compu9ng  
 
 2700  BC  

Mechanical  
1900  AD  

1945  AD   1945  AD  

Electronic  

1960  AD  

1970  AD  
OpBcal  
Semiconductor  

Quantum  

1994  AD  
Timeline  of  computaBonal  implementaBons  

Molecular  

2014  AD   2014  AD   2014  AD   2014  AD  

Semiconductor  compu9ng  -­‐  Silicon  is  only  one  out  of  many  different  semiconductor  
materials.  A  combinaBon  of  the  elements  gallium  and  arsenic  forms  crystals  which  
permit  electrons  to  move  faster  than  in  silicon,  so  that  this  material  is  someBmes  used  
when  extreme  speed  is  important.  The  main  reason  that  silicon  is  used  in  computers  is  
because  it  is  easier,  and  therefore  less  expensive,  to  make  complicated  circuits  out  of  
silicon  than  for  any  other  material.  Computer  circuits  also  require  some  parts  to  be  made  
out  of  insulators  in  addiBon  to  the  parts  that  are  semi  conducBng.  With  silicon,  it  is  easy  
to  make  a  good  insulator  by  adding  some  oxygen  to  produce  silicon  oxide.  The  average  
cost  is  now  much  less  than  one  penny  per  10,000  transistors.    
Op9cal  compu9ng  -­‐  uses  photons  produced  by  lasers  or  diodes  for  computaBon.  Photons  
promise  to  allow  a  higher  bandwidth  than  the  electrons  used  in  convenBonal  computers.  
Photonic  logic  is  the  use  of  photons  (light)  in  logic  gates  (NOT,  AND,  OR,  NAND,  NOR,  XOR,  
XNOR).  Switching  is  obtained  using  nonlinear  opBcal  effects  when  two  or  more  signals  are  
combined.    
Molecular  compu9ng  -­‐  a  form  of  compuBng  which  uses  DNA*,  biochemistry  and  
molecular  biology,  instead  of  the  tradiBonal  silicon-­‐based  computer  technologies.  It  is  in  
its  infancy  and  is  currently  the  subject  of  research.        

*  A  bioengineer  and  geneBcist  at  Harvard’s  Wyss  InsBtute  have  successfully  

stored  5.5  petabits  of  data  —  around  700  terabytes  —  in  a  single  gram  of  
DNA,  smashing  the  previous  DNA  data  density  record  by  a  thousand  Bmes.    
Quantum  compu9ng  -­‐  a  computaBon  device  that  makes  direct  use  of  quantum-­‐mechanical  
phenomena,  such  as  superposiBon  and  entanglement,  to  perform  operaBons  on  data.  
Quantum  compuBng  is  sBll  in  its  infancy  but  experiments  have  been  carried  out  in  which  
quantum  computaBonal  operaBons  were  executed  on  a  very  small  number  of  qubits.  Both  
pracBcal  and  theoreBcal  research  conBnues  to  develop  quantum  computers  for  both  
civilian  and  naBonal  security  purposes,  such  as  cryptanalysis.    
 Advanced Microprocessor System Block Diagram

Color  key:  
Red  =  processor  
Black  =  I/O  
Lt.  gray  =  single  chip  
19  =  contained  on  chip  
Dark  gray  
 Package size

Stacked  memory  package-­‐on-­‐package  (POP)  device  

intro-003

Figure 1-4. Stacked Memory Package on the POP Device

12 x 12 mm

Top memory MCP package

Memory interface

Bottom PBGA547 OMAP4430

POP package 0.4-mm pitch
intro-004

Two types of LPDDR - SDRAM memories are supported in POP package: S4 and S2 with size up to 2GB
and 32-bit data width.
The POP device includes feedthroughs. The feedthroughs are defined from the bottom ball-grid array
(BGA) to the stacked memory. The purpose of some of the feedthroughs is to provide power supply to the
stacked memories.

270 Introduction SWPU231R – July 2010 – Revised March 2011

© 2010–2011, Texas Instruments Incorporated

Quad  Core  14nm  
1.6  GHz  Base  Frequency  
2.4  GHz  Burst  Frequency