A Case Study on Rapid Prototyping of Hardware Systems: the Effect of CAD Tool Capabilities, Design F

1
A Case Study on Rapid Prototyping of Hardware
Systems the Effect of CAD ToolCapabilities,
Design Flows, and Design Styles

• Apostolos Dollas, Kyprianos Papademetriou,
  Euripides Sotiriades
• Dimitrios Theodoropoulos, Iosif Koidis, George
  Vernardos
• Dept. of Electronic and Comp. Eng.
• Technical University of Crete,
• Chania, Greece
• dollas_at_mhl.tuc.gr

RSP June 30, 2004
2
Outline

• History And Motivation
• Case Study Parameters and Algorithm
• Experiment Description
• Results
• Conclusions
• Future Work

3
Evolution of CAD Tools

• A decade ago
• Commercial CAD Tools had severe limitations.
• Design with HDLs and schematic capture demanded
  high degree of detailed knowledge of the
  application as well as the implementation
  technology.
• Reusability was generally limited to the
  designers personal library.
• Design cycle was nonetheless substantially
  shortened.
• Nowadays
• Commercial CAD Tools work very well.
• Designs with HDLs and HLLs, the designer doesnt
  need to know many things about the implementation
  technology.
• Effective and widespread reusability through IP
  Cores

4
Questions To be Answered Today

• Design Behavioral vs. Structural
• New Style of Design (e.g. higher level HLLs) vs.
  Old fashioned (lower level HDLs)
• Vendor Tools Evaluation
• Third Party Tools Evaluation
• Interoperability between Third Party and Vendor
  Tools
• FPGA Vendor Technology

5
Experiment Parameters

• A set of designs with
• Common specifications
• Complete up to post place and route point
• Full timing analysis
• Different implementation vendor technologies
• 5 Engineers with different experience
• Implementation of several versions of the same
  design by each engineer

6
Experiment Parameters (cont.)

• Large number of CAD tools
• Vendor and third party synthesis Tools
• FPGAs and PR CAD Tools from two major FPGA
  vendors
• Simulation with commercial third party CAD Tool
• Most designs in VHDL and one set in Verilog

7
Selected Algorithm

• The selected algorithm is Game of Life (GoL)
• Scalable it tests CAD Tools at small scale vs.
  large scale designs
• Regular the tradeoff between hand placement
  vs. fully automated design can be evaluated
• Understandable designers could have an optimal
  design in terms of resources

8
Description of Experiment

• Three versions
• GL10 10x10 array
• GL20 20x20 array
• GL30 30x30 array
• CAD Tools used
• ST1, ST2 Synthesis Tool of FPGA Vendor A and B
• ST3,ST4 Commercial Synthesis Tool of Vendor C
  and D
• PR1, PR2 Place and Route Tool of FPGA Vendor A
  and B
• SIM Commercial Simulation Tool
• Designers
• D1, D2 7 years of experience
• D3,D4 3 years of experience
• D5 2 years of experience
• FPGA Devices
• FPGAs of Vendor A and B
• 0.13 um, 0.15 um, 0.18 um, 0.22 um

9
FPGA_A_013
FPGA_A_015
FPGA_A_018
FPGA_A_022
10
FPGA_B_013
FPGA_B_015
FPGA_B_018
FPGA_B_022
11
Some Conclusions

• Conclusion (1) We may have similar results
  between behavioral and structural designs for
  some of the tools
• Conclusion (2) Depending on the synthesis tool
  the designer must choose the appropriate design
  style (behavioral or structural)
• Conclusion (3) Compilation of structural designs
  produces results (although the design may
  potentially not fit in the target device) in
  contrast to behavioral design which may not pass
  the compilation of a tool

12
GL10
GL20
GL30
13
More Conclusions

• Conclusion (4) Third-party synthesis tools
  produce much better results with PR2 (place and
  route tool of FPGA Vendor B) than with PR1 (place
  and route tool of FPGA Vendor A)
• Conclusion (5) Vendor-supplied synthesis tool of
  FPGA Vendor A has better results than
  vendor-supplied synthesis tool of FPGA Vendor B.
• Conclusion (6) CAD Tools are not fully reliable
  even for simple designs

14
Some More Conclusions

• Conclusion (7) Third-party synthesis tools
  produce much better results with PR2 of FPGA
  Vendor B than the vendor-supplied synthesis tool
  does. FPGA Vendors A-supplied synthesis tool
  works quite well
• Conclusion (8) Depending on the design style
  (behavioral or structural) the designer must
  choose the appropriate synthesis tool

15
Designer style effects

• More experienced designers use cell-based
• Less experienced trusted CAD Tools
• Behavioral designs produce quite good and many
  times better results than structural with use of
  the proper CAD Tool

16
and final Conclusions

• Conclusion (9) The implementation of designs of
  less experienced engineers may have good results
  when the appropriate synthesis tools are used
• Conclusion (10) It is possible to observe
  substantial differences in the results between
  behavioral designs that are implemented with the
  same CAD tool
• Conclusion (11) It is possible to observe
  substantial differences in the results between
  structural designs that are implemented with the
  same CAD tool

17
Future Work

• Time for Design Cycle and CAD Tools
• Designer vs. CAD Tool at Highly Optimized
  Designs
• CAD Tools for FPGA platforms (e.g. with network
  interfaces, embedded processors, etc.)
• More Complex Designs

18
The End
Thank you!