After years of skepticism over whether the concept could compete with the proprietary tools and intellectual property (IP) that today dominate the market, both hardware and software in open source form are now taking root in the design of electronic hardware.
Electronic hardware design is no stranger to free and open source software, such as the SPICE (Simulation Program with Integrated Circuit Emphasis) analog simulator put into the public domain by researchers at the University of California at Berkeley in the early 1970s. Yet most users will access SPICE through proprietary tools that have each extended the engine in different ways.
But the pendulum has started to swing back towards the use of free and open source tools through a number of initiatives that seek to open up chip design to a wider audience, boost the speed of innovation in specific areas, or break the dominance of proprietary hardware architectures.
Realizing that the cost of custom silicon design was increasingly out of reach for researchers and even suppliers to the defense industry, the U.S. Department of Defense (DoD) Defense Advanced Research Projects Agency (DARPA) sponsored the creation of an open source suite of chip-design tools in 2018 by teams working at a group of U.S.-based universities.
Going under the name OpenROAD, the team focused on building a design flow that could take a hardware description written in the register-transfer level (RTL) language Verilog through to being ready to send to a fab for manufacture in the space of 24 hours, assuming no design errors needed to be corrected.
About eight months after the release of version 2.0 of OpenROAD in late summer 2021, more than 220 designs had been sent for fabrication on a 130nm process provided by Minnesota-based SkyWater Technology, many of them part of a program sponsored by Google to make chip design accessible to a much wider range of users and to try to encourage work on custom silicon not just by hardware designers, but software engineers to accelerate AI applications, with the open-source RISC-V instruction-set architecture becoming an increasingly popular foundation for such projects.
Google is adapting its software build tool Bazel to act as a front-end to a highly automated OpenROAD flow, as well as working on a high-level synthesis tool called XLS that is intended to deliver both software and accelerated hardware for applications. At the end of July, the Department of Defense (DoD) awarded SkyWater $15 million in funding to extend open-source silicon to that foundry's 90nm process.
At the Design Automation Conference (DAC) in San Francisco in July, Peter Gadfort, team lead for silicon technologies at the U.S. Army Combat Capabilities Development Command (DEVCOM), says their ability to use OpenROAD in place of commercial tools has made it easier to explore novel architectures and implement them on advanced processes such as GlobalFoundries' 12nm technology node. Gadfort explained the problem with commercial tools is that their limited licensing periods rarely matched up with project lifecycles. "We often get to act as mini-lawyers," he says. "Open source helps us do research to help our customer, the DoD. The level of integration we are aiming at is just not practical for us in a closed-source environment."
At the same time, Gadfort sees an issue for OpenROAD and projects like it being one of developing a sufficiently large user base. "It is important to grow the user bases and develop sustainable business models. If they are to succeed, they need a viable path to generate income."
Andrew Kahng, distinguished professor of electrical and computer engineering at the University of California San Diego (UCSD), one of the institutions that have coordinated OpenROAD development, points to the dedicated base of supporters that the project has picked up, including Google. "The core team is quite small and will need additional support to realize the complete vision. But in the meantime, we are happy to work with the team that we are able to support – to push the technology forward."
Though chipmakers are unlikely to replace their existing flows built on commercial tools, engineers at chip makers such as Intel and Qualcomm see the possibility to integrate open source tools as they provide a means for collaboration on areas where they do not see existing solutions.
Noel Menezes, director of Intel's strategic CAD laboratories, points to domain-specific languages used to codesign the hardware and software for custom accelerators as an area where open source can make a major contribution. "Open source has to come up with some disruptive ideas that are complementary to commercial tools. The EDA industry continues to solve some big problems and I would not recommend open source [developers] to go there."
The OpenROAD effort may also help foster greater cooperation among chipmakers as the emphasis in tool design shifts toward much greater use of machine learning, a major research interest of Kahng's. A key challenge for the technique in electronic design automation (EDA) is the secrecy around most chip-design projects. However, with a wider user base, it may prove easier than before for the METRICS 2.1 project that UCSD is now running to obtain useful data to kickstart the process and encourage more secretive commercial teams to voluntarily submitt data on how their circuitry translates into finished designs if they see that translate into improvements tool algorithms.
"Ultimately, commercial teams will share according to how METRICS 2.1 helps them address their own design and machine-learning challenges," Kahng says.
That same self-interest may at the same time see commercial developers expand their commitment to open-source projects, in the same way they have taken root in software engineering.
Chris Edwards is a Surrey, U.K.-based writer who reports on electronics, IT, and synthetic biology.
No entries found