At this week’s 59th Design Automation Conference (DAC 2022) in San Francisco, a panel discussion focused on the question: “Is democratization of chip design already happening?” Here, one of the panelists, Vic Kulkarni, chief strategy officer at Si2, presents a summary of the discussion and his own additional notes.
Before we delve into his notes, to set the scene, the panel discussed how far we have come when it comes to democratizing silicon. This is based on the fact that silicon has become ubiquitous, no longer restricted to powerful servers, and silicon content has found its way into many systems. Bespoke silicon is no longer rare, and often a necessary step to enable differentiation. The panel explored whether silicon design and manufacturing is still only accessible to just a few, or what might be done to remove the substantial barrier to entry.
Here are the panel questions (posed by me), and Vic Kulkarni’s notes.
Nitin Dahad: What do we mean by democratization of chip design?
Vic Kulkarni: Plato, the Greek philosopher, wrote during the early 1,500’s that equality without rules brings power-seeking individuals or institutions motivated by personal gain. They can become highly corruptible, and this can eventually lead to tyranny. In this context, democratizationwith membership makes more sense for the world of chip design, where there is a spirit of community, collaboration, member interaction and contribution, and most of all, data security, to easily design and produce chips.
Today the bar for creating innovative bespoke silicon designs has been raised to over $50 million per chip, in addition to creating a software stack to make it successful. This has resulted in an elite class of IP, SoC and silicon-to system designers.
We have to lower the barrier to entry from both financial and technology investment points of view for a larger ecosystem:
- Give access to IP (soft and hard) building blocks
- EDA tools to university researchers, students
- Innovative startups focused on new applications ranging from 5G to 6G, hyperscalers, AI, edge compute, automotive 3D sensors.
In this context democratization with strong guard rails for IP protection, and security for all members. One well proven model is Si2’s OpenAccess, an example of democratization with membership of over 50 companies, including contributors, developers, ranging from foundries, IP, EDA and chip makers. This now contains over a million lines of code.
Nitin Dahad: Is open source a key part of this democratization?
Vic Kulkarni: Open source is one of the key parts of democratization. It is an enabler, since it eliminates the cost of purchasing commercial EDA tools. The work by Professor Kahng and contributors of the OpenRoad project is an excellent example of this. Github, Kafka, and RedHat are good examples. FPGA has an infrastructure that enables democratization. RISC-V is another important example.
However, we can think of other options which EDA companies could offer with well-defined business models for democratization.
- A startup version, perhaps as SaaS
- Tiered pricing based on the complexity of the problem, number of users
- Minimum fee for universities, students, researchers.
We should also bring order to the chaos through the collaborative membership model. Technical contributions from the community, access control, security and traceability for downloads, may be using a blockchain-based approach. An excellent example of this is the emergence of a well-defined ecosystem of apps stores with secure processed data APIs created and supported by the FAANG community.
The value proposition of such APIs is quite clear:
- Enables a developer ecosystem
- Spur innovative applications developed on top of software platforms
- APIs are well documented and provide secure, consistent and flexible interfaces for users.
The outcome of such democratization has seen an impressive growth of apps using platform-specific APIs. It has resulted in an explosive growth of FAANG companies and their platforms. Of course, the knowledge gap between having a nascent idea and converting into working C or Python code for application workloads and translating that through the tape-out process is a big challenge.
Nitin Dahad: Are the traditional EDA vendor business models a help or hindrance to democratization?
Vic Kulkarni: The traditional EDA vendor models are a hindrance to democratization. However, EDA companies are evolving to adapt to the needs of new entrants. Three major players, Synopsys, Cadence and Ansys CDNS have announced ratable SaaS like models for parts of their product portfolio, and it’s reasonable to assume there is more to come.
The recent push towards cloud-based business models certainly enable new players in chip design to get access to a broad set of EDA flows without the need to invest large amounts upfront in tools and compute infrastructure. Silicon Catalyst is an excellent conduit for enabling startups with their in-kind partner program.
Nitin Dahad: What about skills – what needs to be done to make it easier for non-specialists to exploit the democratization of silicon?
Vic Kulkarni: Before we get to democratization, there is already a tremendous talent deficit, due to the recent growth in semiconductor system designs using bespoke silicon for application-specific system verticals. As mentioned, we now have only the elite class of IP, SoC and silicon to system designers.Enabling the necessary workforce requires private-public partnerships – such as the government funding in excess of $450 million per year for Purdue University in the U.S. to massively increase the number of graduating students in multiple disciplines.
We also see that non-traditional education mechanisms can come into play too. Today, there are many online resources available. However, if this level of training is needed, then democratization won’t happen. We need better tools that can take ideas to silicon in a fully automated way – for example, “compiling” from a high-level behavioral language, without requiring the knowledge of layout, IP design, fab process, and so on.
More focus on out-of-box solution, whether they are focused on a market vertical (e.g. automotive, AI/HPC, memory) or a technology horizontal (energy efficiency, multi-die systems, thermal). AI/ML-based automation is another area that can make life easier for non-specialists.
Nitin Dahad: What is the one thing that is most important to you with regard to where we are in this process, and the one thing that you’d like the audience to go away and think about?
Vic Kulkarni: My view is that the next wave of innovation will come from a system-first mindset for the silicon-to-system innovation with ecosystem collaboration. Dr. Mallik Tatipamula, CTO, Ericsson Silicon Valley, recently shared an important of vision of a multi-dimensional ecosystem mindset to make the transition from 5G to 6G a realizable goal within next 5 years.
What I’d like the audience to think about are things like:
- How do we overcome this barrier of talent shortage? What are the creative ways in which this can be done?
- What will be a more practical approach? For example, programmable silicon or systems. FPGA is one form, where it’s already been manufactured.
- In the edge IoT space, many are producing kits that allow one to customize the system to meet their needs. This is a more likely middle ground.
- Chip design in the cloud? Now you can have pay-as-you go EDA
- The future of cloud convergence and the three P’s
- Finding the triggers for open-source hardware success