October 06, 2020
Open source EDA solutions have recently been put back on the agenda due to the semiconductor industry’s interest in open source hardware. However, there is no definite answer to these questions about whether there is enough motivation for open source EDA this time to advance its success or the scale of development.
One of the main promoters of this open source EDA is the US Department of Defense Advanced Research Projects Agency (DARPA for short), which has led a number of projects to reduce chip design costs, one of which is research on advanced packaging technology, and the other It's about chip security. The reason why DARPA has invested in these projects is that it hopes to find ways to make chips cheaper and less risky through the research of millions of existing chip designs.
DARPA ultimately hopes that these research results will enable the U.S. military to obtain suitable chips to deal with the increasingly complex communications, reconnaissance and weapon systems. In order to achieve this goal, they are also actively lobbying other organizations to get more support by bringing them the same benefits.
DARPA-related project manager Andreas Olofsson once mentioned: "Now it costs 100 million US dollars to design and produce a complete SoC. Among them, EDA tools are not the main factor driving up the cost, nor are the tape-out costs. In fact, most of them are spent on engineering development. For example, chip design and component development of related software stacks. We must find ways to reduce engineering complexity and cost. How to do this, I can design IP through automation and reuse, not just reuse components , But the reuse of its entire subsystem, which can make the design process cheaper and easier to use."
Olofsson left his position as CEO of Adapteva in 2017 and came to DARPA to lead the two projects that impact the EDA industry, Intelligent Design of Electronics Assets (IDEA) and PoshOpenSource Hardware (POSH). Its goal is to help us train applications accelerated by machine learning by using existing chip designs as a data source, so that we can better debug, verify, and automate placement and routing. This IC database will include more than 5 million components currently in circulation, and can standardize model names to replace non-standard design descriptions that only exist in data tables, so that we can use the data in a constraint-based optimization system.
Olofsson mentioned, "There are many engineers in the U.S. Department of Defense who can skillfully write good-quality Verilog code and run it on FPGA, but they do not have an internal team to convert it into ASIC chips. We now want to lower this threshold. The lower the better."
Both IDEA and POSH are part of DARPA’s $1.5 billion ElectronicResurgenceIniTIaTIve program (ERI). ERI is also advancing chiplets and projects dedicated to simplifying packaging and chip design. The final realization of these projects depends on For open source software.
"We need to be more aware of the impact of open source, because open source means more communication costs," Olofsson said, "This is a very challenging problem, because it means that many teams will join the development and a large number of Cross-team cooperation. But we have seen many examples that prove that open source is actually very effective. The latest example is its role in the development of machine learning. It is an order of magnitude impact to promote innovation and cooperation. At the same time, when we When carrying out this kind of industry-disruptive development work that requires multi-team collaboration and generating new components, open source is also the only practical method under the academic research setting. Here I strongly advocate the use of a tolerant open source license. It means that anyone can use these codes and build their own proprietary solutions, while only giving back their basic code to the community, so this will be a very gentle open source license."
Open source EDA will not appear immediately
The price and flexibility advantages brought by open source have attracted the attention of many companies. Even though many of them are not currently using open source solutions, these companies now only choose to use the best solution for them, regardless of whether they are open source or not, there are sufficient reasons behind them.
Linley Gwenapp, the chief analyst of Linley Group, said in an interview, "EDA companies tend to focus only on their major customers, such as Broadcom, Qualcomm and Intel. They all have a large market share and have many different SoC designs. , And these big companies themselves have the motivation to find the best design solutions and processes. If they don’t, they may lose out to competitors because their products don’t work or are too expensive, so it’s reasonable for them to work closely with EDA companies. Of course, they hope to minimize the engineering investment, but they can’t pay the price of inferior products. For these large companies, the investment in the early engineering development is compared with the later sales of these products on the market. In terms of return, it is not a particularly large amount."
Oloffson believes that the status quo is that the resources of the entire industry are tilting toward these few large companies to help them build the most complex processors, and expect high sales and extremely low chip error rates. However, as chips enter some markets with high safety requirements, such as the field of unmanned vehicles, their requirements for reliability are also higher, making it difficult for the EDA industry to maintain this cooperative relationship. And despite the industry's criticism that commercial EDA tools are difficult to master, EDA companies have been actively working to simplify the design process for their customers and find ways to shorten the time to market.
In addition, more manufacturers are not developing chips that use the most advanced nodes, which means that they do not need the most advanced design tools. FPGA vendors have been providing relatively simple tools developed by themselves for many years. Due to AI/machine learning, more and more people have begun to use GPUs, and these engineers have been using solutions that everyone is already familiar with.
Upheaval in the industry is about to begin
Almost no one in the industry would disagree that today's chip design is too complicated and expensive. The question now is how we will solve this problem, especially when the market becomes more segmented and flat, and structural changes will impact or even replace the traditional model of growth.
NXP Semiconductors Software R&D Department VP Robert Oshana said in an interview, "Open source is still a relatively new concept for the hardware department, but it has been developing rapidly. We can see that the development of open source in the software industry is also affecting the hardware industry.
He pointed out that open source does not just mean RISC-V. "This will have a disruptive impact on the industry, but we are likely to find that just like open source for the software industry, it will bring many benefits to the hardware industry, so (the hardware industry) must change to adapt to this. "
Others are not so sure. Mentor Executive Vice President Joe Sawicki said, “If we know what positive impact open source will bring to us, then we will definitely invest more internally in related research and development. It can be seen that such projects will be very exciting. The potential of excitement, but the status quo is that we are not sure to achieve the ultimate goal, nor are we sure what benefits will ultimately be obtained."
Open source itself also brings certain risks. Synopsys senior security information expert Taylor Armerding published an article on his company blog earlier this year on how to deal with serious vulnerabilities. "After nearly a decade of development, SPDX (SoftwarePackageDataExchange) will add hooks for handling open source vulnerability data."
All large EDA companies and industry-leading chip companies are active contributors to the ERI project. In fact, Cadence, Synopsys, Mentor, NXP (NXP), Intel, IBM, Qualcomm, ARM, Nvidia, AnalogPhotonics, SRI International, AppliedMaterials, etc.
Many companies not only simply participate, but also major contributors who need a large amount of open source development resources such as POSH and IDEA. Nevertheless, whether open source EDA can be accepted by the industry still depends on whether it can handle the design process more efficiently, and at the same time, it cannot break existing rules. In addition, it also depends on whether it can extract effective information from its own million-level design library, and use it to detect and unearth existing design errors in real time.
Olofsson added, "For DARPA, the question we are studying is how complex the design can be handled, and what the final effect is. Smart chip development projects have a variety of design techniques, and they apply them to the company. Chip products and finally tape out. These skills only exist in engineers and not in EDA tools. We hope that by advancing these DARPA projects, we can extract these experience skills from existing designs and apply them to new tools In the new design, the labor cost of project handover is saved."
In this regard, further studies are needed to confirm its feasibility. But at least for now, there is enough budget to support the exploration of open source EDA tools.