A few weeks ago, we signed a term sheet with Alpha Design, an early-stage company developing AI to assist in the development and testing of digital chips. The company was founded by William Wang, an accomplished professor at UCSB who has led the university’s AI efforts over the past several years. Semiconductor designers use a category of programming languages known as Hardware Description Languages (HDLs), the most popular of which is Verilog. While large language models (LLMs) have been effective in assisting with programming tasks, their performance is correlated to the popularity of each language, especially in the open-source community which provides much of the training data. This means GPT-4 can deliver best-in-class results in Python and JavaScript but becomes less competitive as we move into less common languages. Verilog is an example of a less common programming language. Therefore, starting with an open-source LLM and improving it to outperform GPT-4 or similar models presents an opportunity to develop a differentiated product for the semiconductor industry.
As some of you might have speculated, betting against the continuing improvement of OpenAI and other top AI labs is risky. We are aware of this risk, which is why we look for defensibility in the long term. We have viewpoints on various forms of AI defensibility, one of which concerns complex feedback loops. With a typical programming language, it’s easy to test whether the program behaves as expected by simply executing the code, often in the browser. So, if I ask GPT-4 to write a function that calculates the Fibonacci series up to the Nth number, I can test it by asking to compute the 10th term and comparing it with the known value of 55. However, testing HDL is more complicated—the circuit has to be simulated under certain conditions. If the circuit is highly complex, these simulations can take much time and require substantial computation. Furthermore, many simulation products are closed-source and require a commercial relationship.
And this is just to test the logical behavior of the circuit. The next step is to consider the actual physics of the manufacturing process, perform additional tests to understand power utilization, interference, etc., and iterate once again. All of this has to happen before a single silicon wafer is turned into chips, which means improvements in the development lifecycle of semiconductors can have dramatic implications for time to market.
The global integrated circuits (IC) market was $490B in 2022, and Nvidia’s market cap has grown close to 800% since. The global electronic design automation (EDA) tools market is currently $17.7B with an expected CAGR of 8.46%, with players such as Cadence and Synopsis. There is an interesting timing factor in this market, given that a lot of the economic activity is linked to the Asia-Pacific region, at the same time as the USA and its allies aim to achieve semiconductor independence. For example, the USA approved the CHIPS Act with strong bipartisan support to allocate $280B to boost domestic research and development of semiconductors. We believe this will provide tailwinds to startups such as Alpha Design.
More about William Wang: he is the Director of UC Santa Barbara's Natural Language Processing group and Center for Responsible Machine Learning, is also the Duncan and Suzanne Mellichamp Professor of Artificial Intelligence and Designs in the Department of Computer Science at UC Santa Barbara. He holds a PhD from Carnegie Mellon University's School of Computer Science. William's research spans various AI fields, including statistical relational learning, information extraction, computational social science, dialogue and generation, and vision, with over 100 published papers and numerous accolades, including best paper awards, DARPA Young Faculty Award, IEEE AI's 10 to Watch Award, NSF CAREER Award, and the British Computer Society. He has received multiple faculty research awards from Google, IBM, Facebook, Amazon AWS, JP Morgan Chase, and Adobe.
We are very excited about this investment. Ivan has known Professor Wang for many years and has been waiting for the opportunity to collaborate on building a business. Ivan Bercovich and John Bowers, another academic-entrepreneur success story, will be sitting on the board.