Is “designing AI chips with AI” feasible?
This article is based on an interview undertaken by FusionMedium’s technology online media, TechOrange, and published with permission:
Semiconductor chips have become the core for driving innovation in the edge computing of tomorrow. Smartphones were the world’s first stage of entry into the Era of Intelligence. As reliance on electronic products continues to grow, how to design higher performing and cost-effective chips, and to do so in a shorter time, has become a key challenge for IC designers.
Synopsys is an S&P 500 company and has a long history of being a global leader in IC electronic design automation (EDA) and IC interface IP. The company is dedicated to providing the best “Silicon to Software” solutions.
EDA has greatly increased the speed at which technology evolves, while the number of transistors deployed in a chip has risen sharply
With the simple and rapid dissemination of knowledge, human technology has evolved so much that the words that used to be printed by the Gutenberg printing press have now become digital zeros and ones, logical gates, and tools to miniaturize these technologies in tiny silicon chips, and that tool is Electronic Design Automation (EDA).
The emergence of EDA has greatly increased the speed at which technology is developed. Li pointed out that semiconductor technology became a key player in all kinds of electronic devices and systems, although IC circuits were still designed manually at that time.
Li recalled the early 1990s when there were only 12 million transistors in telephone chipsets and engineers were still drawing circuits by hand, taking a year and a half to complete the design of a single chip. “The number of transistors in cell phone chips on the market today runs between 1.2 billion and 2 billion, and each transistor circuit has to be connected correctly for the phone to work, which is virtually impossible were the old design method were still in use. EDA makes this possible. “
EDA design involves three processes to meet the needs for higher performance, lower cost and faster time-to-market
The first step of EDA is to describe the circuit, and the second step is to build a circuit model, simulate operation, analyze feasibility, optimize performance, and finally reach automation.
This process is similar to the current AI machine learning algorithm, which also collects a large amount of data, simulates it into a neural-like network, and then trains the model so that the system can make inferences and predict behavior, and design corresponding actions.
Li continued his description, “AI technology can be applied to EDA in every aspect of IC design, from specification, functional and circuit design, to real-world verification, to IC production and testing, in a way that human beings cannot. The key indicator of IC design is PPA (Power, Performance, Area), which means less power, higher performance and more transistors packed into the same geometry.
In addition to PPA, various considerations such as information security and stability, among others, must now be included, which makes IC design more and more complex.
The era of “designing AI chips with AI” is coming
As a global EDA leader, Synopsys launched an AI-enabled EDA platform in 2020 and joined Taiwan’s Industrial Technology Research Institute (ITRI) to establish the AI Chip Design Lab. The firm then launched the public AI system-on-chip (SoC), to assist IC designers in shortening the development timeline. “If we continued to use the old design method, it would take 100 engineers more than three years to complete the design of an AI accelerator for social networking sites. After leveraging the AI SoC, the Taiwan start-up team needed only 30 engineers and completed the product design within a year and a half.”
The AI on Chip Industry Cooperation Strategic Alliance established by the Smart Electronics Industry Promotion Office (SIPO) of the Industrial Development Bureau, connects the upstream and downstream supply chains of global industries and assists semiconductor, AI, and IoT manufacturers in establishing international partnerships, giving Taiwan’s manufacturers an opportunity to explore innovative business opportunities and enhance international competitiveness.
Lastly, Li said that the global industry relies more and more heavily on ICs as we enter the Era of Intelligence. Statistics show that future market demand for IC design engineers will be 1000 times higher than it is now. To meet this demand, AI is imperative. The era of designing AI chips with AI has come and IC designers must ready themselves for this trend in order to remain competitive in the industrial environment of tomorrow.”