‘End-to-end openness potentially lowers the barriers to getting ideas turned into silicon’

The director of the Microelectronics Design Center at ETH Zurich, Frank Kagan Gürkaynak is a vocal advocate for open hardware, including through the PULP Platform , a collaboration between ETH Zürich and the University of Bologna (Alma Mater Studiorum – Università di Bologna) . With a significant portfolio of tapeouts under its belt, PULP has been pioneering open hardware in Europe and its designs have been used in several commercial products. HiPEAC caught up with Frank to learn more about why openness is central to his hardware philosophy.

What does ‘end-to-end openness’ refer to in hardware development? Why is it important?

End-to-end openness involves open-source principles at all levels of the design of integrated circuits (ICs). Everything in the design process should be accessible to anyone, from the manufacturer’s process design kit (PDK), which contains all the information allowing an IC to be reliably manufactured in a particular technology, to the electronic design automation (EDA) tools used to transform a design idea into a manufacturable circuit, to the description of the design itself.

By dramatically increasing accessibility to all aspects of IC design, teaching and training can be made available to a broader audience, which has the potential to significantly lower the barriers to getting ideas turned into silicon. In addition, there are several applications where transparency is of the utmost importance, and having this level of access to all design stages supports independent audits. Lastly, it removes black boxes in the design process and allows innovation at all levels of the IC design flow.

OK, great! Let’s get started on my next open-hardware idea…

Not so fast. IC design is still a costly endeavour with very high non-recurring engineering (NRE) costs that can only be compensated with volume. Openness lowers the barriers but doesn’t turn the cost structure upside down.

Besides, open-source EDA is still in its infancy: it’s possible to make viable designs, but there is still a gap between open-source EDA and commercial solutions in terms of the features, the usability and the quality of the results, which is to be expected. Currently, designs in mature technologies in the 180nm-130nm would be feasible with EDA tools; some more work is required to move to designs in the 65nm-28nm range, and there is still a gap for the most advanced nodes in 10nm and under.

The important part is the open PDKs being made available. At the moment these are limited to mature technologies, such as the 130nm from the Leibniz Institute of High Performance Microelectronics (IHP) in Germany. We are hoping more will become available. A technology in the 65nm-28nm range, in particular, would allow many industrial designs to be realized efficiently.

Open-source designs, on the other hand, don’t have that many barriers; designs originating from the PULP team have been implemented in practically all technologies.

When we last spoke [see HiPEACinfo 66 p.23], you mentioned that the EDA tools market was dominated by proprietary tools. Is this changing?

This still holds true: at least 99% of the market uses proprietary tools. However, EDA design tools allow design flows to be created with a mix of tools from different vendors, both proprietary and open source. While I don’t expect a wholesale change from completely proprietary to completely open tools, I do expect a gradual penetration of tools from the open-source community into design flows.

One important difference is that, while proprietary tools’ licensing fees place a limit on the number of concurrent users / runs you can have, open-source tools can be used without limitation. This makes open-source tools very attractive for automated tasks such as those in continuous integration and early evaluation, where the design is still being refined.

In these discussions, people often think of high-end designs like machine-learning accelerators destined for data centres. However, this market (while exciting) constitutes a relatively small proportion of ICs designed: you also need chips that end up in refrigerators, your computer mouse, multifunction displays, and many others. For these applications, performance is not as important as it would be for a high-end product. You can afford to have a design that’s a bit larger, a bit slower, but it would help if you could get started right away without negotiating licensing agreements with an EDA vendor, especially if it’s not 100% sure that you will actually commit to an IC design process.

For this reason, I expect that the availability of open-source tools will increase the number of people who get some familiarity with design flows, increasing innovation in this area. This could, eventually, even lead to more business for commercial EDA vendors, as more people with more experience venture to more ambitious projects after their initial foray into IC design on open-source EDA tools and technologies.

All right, I’m ready to embark on my first design. What resources are available for people who want to get involved in this area?

At ETH Zurich, we have a long tradition of IC design courses, dating back to 1986. From spring 2025, this course, which has led to 100s of ASICs being designed and manufactured, will use mostly open-source tools, making exercises and lecture material available in the process. We have already managed to tape out a larger, Linux-capable, RISC-V-based system based on our Cheshire platform in IHP130 technology recently, and we will use the experience from this tapeout to adapt our course material.

In parallel, Tiny Tapeout continues to support completely open-source designs for everyone and they are in the process of adapting their tapeouts to the IHP130 process and planning a run in middle of 2025. The TinyTapeout approach allows enthusiasts, students at all levels to get acquainted with the IC design process and receive chips that include their design (together with many other open designs) at very affordable prices.

Are there any downsides to open hardware?

Some people take an ‘all or nothing’ approach, and point out that open-source EDA tools can’t be used to tape out a high-end graphics processing unit (GPU), for example. Comparisons so far have also shown that, so far, open-source EDA tools don’t deliver the same quality of results as proprietary tools, although the difference is less than some people believe. Others are concerned about the risks involved in relying on ‘free’ tools that do not come with professional support for their IC projects that costs millions.

These are all valid points, but they do not really diminish of open-source EDA tools. There are many designs and projects that can be realized with open-source tools, and that are completely sufficient for teaching and training purposes. The tools are also getting progressively better; recently I experienced this first hand, where within a six-month period the logic synthesis results progressively improved to produce circuits 1.6x smaller and 2.3x faster than they were at the beginning.

The other discussion I hear centres around national sovereignty which either sees open source as a threat (allowing certain geographies access to technologies that could otherwise be limited), or a solution (achieving independence from companies from certain geographies). One needs to realize that open source is open for everyone, and not limited to political boundaries. I do not think we are having the same discussion over GCC or LLVM, so I think these discussions will also with time disappear.

Can you give any examples of successful industry integration of open hardware solutions?

One of the first things you end up learning when working with open source is that you need to get used to 'let it go'. We use a permissive open-source licence based on the popular Apache v2.0 licence that allows anyone to use it as they see fit, make commercial products without necessarily letting us know. We have anecdotal evidence of where our open-source designs were used in industry, but whether this is in the 100s or 100s of millions of products sold, we do not know.