From Synopsys to Google, New EDA Tools Apply Advanced AI to IC Design

September 11, 2023 by Jake Hertz

EDA tools now employ the newest advances in AI to help designers accelerate the IC development process.

For years, EDA companies have claimed “artificial intelligence” features in their IC design tools. In the past year, however, generative AI has undergone a dramatic evolution with platforms like ChatGPT, causing some designers to question whether previous EDA features still count as AI by today's standards. 

Synopsys aims to keep pace with this accelerating field by unveiling a new extension to its EDA suite. This announcement follows the release of Google’s DeepMind, which uses AI to accelerate its in-house chip designs. Both of these announcements indicate how advanced machine learning algorithms are shaping IC development and how they might be used as a tool for designers in such fields. 


The reinforcement learning cycle's reinforcement learning cycle. Image used courtesy of Synopsys Aids in Comprehensive IC Development

Synopsys describes its new extension as an AI-driven analytics tool designed to span the entire integrated circuit development process, from initial design to manufacturing and testing. To this end, the Synopsys EDA Data Analytics solution offers several features that set it apart. 

First, it provides comprehensive data aggregation capabilities, pulling in data from various stages of IC design, testing, and manufacturing. This gives designers a holistic view of the entire chip development lifecycle. The tool incorporates “intelligence-guided” debugging and optimization, which not only speeds up design closure but also minimizes project risks. This is particularly crucial in an industry where time to market can be a make-or-break factor.


Synopsys'  Test Space Optimization

Synopsys' Test Space Optimization ( solution, part of the new extension, achieved optimal pattern count in tests. Image used courtesy of Synopsys


Another standout feature of the extension is its focus on fabrication yield. This tool is designed to improve fab yield for faster ramp-up and more efficient high-volume manufacturing. Additionally, the tool can uncover silicon data outliers across the semiconductor supply chain, thereby improving chip quality, yield, and throughput.

Synopsys says the new tools can also uncover new opportunities in power, performance, and area (PPA). By leveraging advanced AI algorithms, the tool can analyze magnitudes of heterogeneous, multi-domain data to accelerate root-cause analysis. 


Google Uses DeepMind for Internal Chip Design

The news from Synopsys comes on the heels of a similar announcement from Google's parent company, Alphabet.

Recently, the group announced that it would be leveraging Google's DeepMind for AI-assisted chip design for use in its data centers. DeepMind uses a concept known as “circuit neural networks” to treat a circuit as if it were a neural network, turning edges into wires and nodes into logic gates. 


A graphical depiction of circuit neural networks

A graphical depiction of circuit neural networks. Image used courtesy of Google DeepMind

Then, using classical AI techniques like simulated annealing, DeepMind searches for the most efficient configurations, looking many steps into the future to improve circuit design. Utilizing advanced AI models like AlphaZero and MuZero, which are based on reinforcement learning, DeepMind has achieved "superhuman performance" in various circuit-design tasks. 


AI Helps Accelerate IC Design Process

While both Synopsys and Google's DeepMind are leveraging artificial intelligence to revolutionize chip design, their approaches and focus areas are distinct. 

Synopsys' newly announced solution is part of its broader EDA suite, which aims to provide designers with an end-to-end, comprehensive toolset for the entire IC chip development lifecycle. These tools aggregate and analyze data across multiple domains to enable intelligent decision-making, speed up design closure, and improve fabrication yield. 

DeepMind, on the other hand, takes a more specialized approach. It employs advanced AI models to tackle specific optimization problems within chip design. While highly effective, this approach is more narrow in scope, focusing on individual aspects of the chip design process rather than offering a comprehensive, full-stack solution. Unlike Synopsys’ tool, DeepMind’s AI is only for the internal optimization of Google’s hardware. 



Featured image (modified) used courtesy of Synopsys