From F-Si wiki
Jump to navigation Jump to search

FOSS & business

Most projects are sponsored by

  • EDA vendors
  • Google
  • Chip manufacturers

They are not really "free":

  • Closed communities (Si2)
  • Non-permissive licenses (e.g. MIT)

What do you get with those?

  • Bad taste:
    • You're not steering, just surfing
    • You don't know their agenda: example Si2 -> market Cadence platform, Google(!?)
  • Benefits:
    • Establishes de-facto standards (e.g. RISC-V)
    • Creates some momentum -> third party support (e.g. compilers, architecture platform)
    • Prevents diversification -> focus on one solution if the sponsor is big enough

Real FOSS:

  • Can provide the benefits when properly steered (which is hard)
  • Everyone can engage or fork to regain the "steering wheel"

Free PDK's

FreePDK45 or similar

  • Not entirely free
  • Quality difficult to assess

Real Free PDKs

  • For small-scale production
  • To harvest the contributions from open source community
    • e.g. Open source libraries, DRC, LVS, RCX ...
  • To establish a market for components and/or open source community
    • Open source building blocks (see below)
  • To foster evolution of standards in terms of tech platforms
    • Manufacturing flexibility
    • Specialization of manufacturers in terms of quantify/small scale production, quality, conformance, add-on features, integration, packaging options, temperature range, power options or NV memory options ...
    • Expect a "land slide" effect as soon as one manufacturer starts opening their PDK
    • Fosters design skill deployment / expert development / human resources market

Beside "real free PDK": establish a "state of the art PDK" for a certain node

  • "benchmark design" which allows assessment of compliance
  • "xxx compatible" quality label for a process?
  • As a marketing helper

Advantage over FPGA implementations?

  • Power
  • Performance
  • Analogue designs possible

On the other hand:

  • no benefit from high-integrated IP blocks (as for FPAs)

Free building blocks (as compared to "Intellectual property")

  • eFabless component market is a good idea
  • Like: memories, analog building blocks, pads, logic cells or device concepts, sensors, ...
  • Provide a new concept or an implementation of a known concept of a specific platform
    • Multiple levels of implementation: abstract - logical - physical
  • For example standard cell: abstract (logic model), logical (transisor netlist, design parameters, scaling rules, testbench, constraints, parameter tables, measurement specifications), physical (layout, timing model ...)
  • Foster a generalization approach
    • E.g. have to be adaptable to different technologies
  • Need an open PDK to enter the physical level or even logical level for a proof of concept
  • Closed PDKs make prohibit open development on physical or even logical level
    • For example: SRAM development
  • Patent issues:
    • Less risk with free building blocks? -> Risk assessment possible, generates "prior art"
  • Export control:
    • Need an option to withdraw building blocks or mark them as "dangerous"?
    • Is this a path how "open source" might become dangerous???

Activity proposals

  • Compatibility assessment design (self-testing/benchmarking vs. Modelling)
    • There is a connection to PDK development efforts!!!
  • "first principles" PDK possible?
    • Start with published information -> possible to derive a basic PDK which is "good enough"?