TDEB Top Bar

Trending
Semiconductor Industry News — The Digital Electronics Blog
Follow Us

TDEB Masthead

The Semiconductor Hub

Semiconductors.  Chips and Systems.  Innovation.

Indexed by leading Search Engines, AI and referenced by Engineers worldwide 500+ Articles 10M+ Pageviews 30+ Reports 50+ Citations

Developing Silicon IP with Open Source Tools

mg
by
3
Share Facebook X LinkedIn WhatsApp Email

The electronic design automation (EDA) tool industry is big business, and commercial licenses are extremely expensive. Open standards have driven many proprietary EDA technologies to be publicly released as free/libre open source software (F/LOSS) and some have become IEEE standards. In this article (Technology Innovation Management Review, May 2011), author Arthur Low reviews the history of key advances in ICs and EDA tools. The common theme presented in this article for the driver of technology innovation is the requirement to develop the most advanced microprocessor possible.

Today, a low-cost, high-value-added business model can efficiently serve the market for IC subsystems licensed as intellectual property (silicon IP) in the form of compilable source code. Alternatively, for larger SoC designs, engineering budgets can be shifted from the purchase of a relatively small number of high-cost EDA tool licenses to open source EDA technologies that can be run on massive compute-server farms. The two business models are not theoretical, but realistic. The author explains how his company (Crack Semiconductor) developed commercially successful cryptographic silicon IP using entirely open source EDA technologies and how another company (SiCortex) pushed the limits of IC design and open source EDA tools by simulating and verifying a massively parallel supercomputer.

The idea in plain words

Silicon IP is a block of chip design, such as a processor or a crypto engine, that one company writes and another company puts into its own chip. It is usually delivered as source code in a hardware description language (HDL) such as Verilog. Before it can be sold, the block must be simulated and verified, and that has traditionally meant buying commercial simulators.

Low's argument is that a small IP company does not have to start by spending heavily on tool licenses. Open source tools can turn Verilog into fast C++ models, and ordinary Linux servers can run those models in large numbers. That keeps costs low, and in his view lets founders avoid giving away equity to pay for tools.

Why it matters

  • Cost per seat. The article gives about $25,000 as the list price of a commercial Verilog simulator. For a startup, a few seats can be a large share of early funding.
  • Scale. A license-free simulator can run on hundreds of machines at once, so regression runs finish faster.
  • Control. With the source available, a team can adapt the tool to unusual verification needs.

How the two companies did it, in outline

Crack Semiconductor wrote an RSA public-key security processor in Verilog, converted it to a C++ model with the open source tool Verilator, and tested it from a SystemC testbench, with a free calculator program working out the expected answers.

SiCortex built a parallel supercomputer from many identical processor chips and verified the chip by running Verilator models across a large farm of Linux servers, tracking coverage with an in-house open source tool.

Deep dive: the numbers and details

This part holds the history, tool flow and figures from Low's article. Skip it if the overview is all you need.

The history the article reviews

WhenAdvance, as described in the article
1965Gordon Moore, then at Fairchild Semiconductor, publishes the observation in Electronics (April 19, 1965) later named Moore's Law.
1968 to 1971Moore and Robert Noyce found Intel in 1968; in 1971 Intel produces the first microprocessor, with 2,300 transistors.
Mid-1980sHDLs replace logic diagrams. A US Air Force requirement for a complete functional description leads to the open VHDL specification; Verilog is later released to an independent organization. Both become IEEE standards.
1980s to 1990sLarge companies build in-house tools, for example IBM's Einstimer timing checker and Bell-Northern Research's Funsim simulator.
1994 to 1998DEC develops Verilator to co-simulate C software and Verilog hardware for the Alpha project, and releases the source in 1998. Wilson Snyder has maintained it since 2001.
Mid-1990s onwardCommercial EDA tools let fabless companies design ASICs for foundries; ARM pioneers licensing processor designs as IP, with over 10 billion ARM processors licensed by 2008.

The article also notes that Intel's microprocessor transistor count went from 2,300 to 820 million in 35 years, which it equates to a 2.07-fold increase every two years.

The Crack Semiconductor flow

  1. Write the RSA processor in Verilog. It multiplies numbers of 1024 bits and larger.
  2. Convert ("verilate") the Verilog to a SystemC model with Verilator and compile it with G++.
  3. In a SystemC testbench, generate random operands. Call GNU bc, an arbitrary-precision calculator, to precompute the system constants and expected results.
  4. Read those results from a file, program the virtual processor and run the simulation.
  5. When a result is wrong, use bc scripts to compute intermediate values and compare them with internal observation points such as the 32-bit multiplier.
  6. For delivery, use free Xilinx synthesis tools to convert the soft core to the client's technology-specific format.

The client was a large European company using the processor for banking security transactions.

The SiCortex numbers

ItemFigure in the article
System972-node parallel supercomputer
Node64-bit RISC processor chip of about 200 million transistors
Simulation farmUp to 400 Linux compute servers running in parallel
Coverage trackingCovVise, built on Linux, Apache, MySQL and Perl/PHP

Worked example: the license arithmetic

Using the article's own figures, and assuming one simulator license per server:

Servers running simulation at once  = 400
List price per commercial seat      = about $25,000
License cost at list price          = 400 x $25,000 = about $10,000,000

Real contracts are negotiated and rarely priced per server at list, so treat this as an upper bound. It still shows why the article argues that a team running hundreds of parallel jobs gains most from a license-free simulator.

Limits the article admits

  • Front-end design of silicon IP still needs specialized problem solving, and the article says commercial tools often fall short there too.
  • The open source flow covers simulation and verification. Synthesis for the Crack deliverable used free vendor tools from Xilinx, not open source synthesis.
  • The article names NXP among the companies using these tools and says they are valued for quality, not only for being free.

Further reading

Share Facebook X LinkedIn WhatsApp Email

Post a Comment

3 Comments

Your comments will be moderated before it appears here.

  1. Such a informative. . . thanks for sharing with others . .Highly benefited....keep posting :)

    ReplyDelete
  2. Thank you for the link! It was an interesting read!

    ReplyDelete
  3. Component or end item comprising of a number of parts or sub assemblies put together to perform a specific function. It is amazing information but i think when a post is really good.
    Electronics Assembly

    ReplyDelete
Post a Comment

#buttons=(Ok, Go it!) #days=(20)

Our website uses cookies to enhance your experience. Learn more
Ok, Go it!