Design processes are evolving rapidly, and their use will enable the highly optimized ICs, PCBs and systems that we need to keep global innovation on track. Today’s efforts to apply analysis much earlier in the design exploration and validation process are already enabling complex multiphysics analyses and co-optimization across domains. However, increasing design complexity means we may soon need to move beyond such in-design analysis—to processes enabled by machine learning (ML) and AI.
This may sound like a reach, but ML techniques are clearly very powerful, if applied intelligently, and the one thing that the electronics industry is never short of is design data. Surely there must be a thoughtful way to bring them together.
Today’s process for getting an idea from design to signed-off implementation is based on a series of steps: produce a design, simulate it to see whether it meets the criteria, and if not, adjust and iterate. This works but is constantly challenged by the combinatorial impact of increasing design complexity. We are also finding that ICs, boards, and systems must be developed as one entity to be truly optimized, further increasing complexity. There’s yet another challenge—the physical constraints that apply to one aspect of the design are increasingly affecting other aspects of it as well.
The next step from this “design then simulate” approach is for EDA vendors to make design analysis capabilities available earlier in the flow—what is generically known as a “shift left” and what we call “in-design analysis.” This gives designers the ability to do more systemic design and co-optimization, such as analyzing a signal’s integrity as it crosses a SerDes communications channel, from the transmitter IC silicon, through its package, across a PCB, to the package of a receiver IC, and finally into its silicon.
In-design analysis enables designers to think more systemically, safe in the knowledge that the tools will bring forward cross-domain constraint violations and highlight optimization opportunities. The challenge then becomes to explore this expanded design space efficiently, especially if designers who usually work in one domain (ICs, say, or PCBs) aren’t familiar with optimizing in the other domains.
This is where ML techniques come into play. Generative ML techniques can produce design options, based on a model that is initially trained on the physics of multiple candidate designs, such as vias connecting layers in an IC. As each candidate is simulated, the resultant data is passed back to the model, which is used for reinforcement learning. The model is updated so it can generate better candidate designs, which in turn drives further evolution of the model. This approach, which Cadence has implemented in its Optimality Intelligent System Explorer tool, helps designers address the “tyranny of choice” they face in vastly expanded system design spaces.
The next step in applying ML to electronics design is likely to involve applying to the vast design datasets that designers and tool providers hold, to extract patterns that represent cues for design success or warnings of failure. Design then becomes a collaborative effort, in which a designer and an AI “co-pilot” work together to explore a design space that has been constrained by the patterns revealed from ML analysis. In this scenario, early design exploration may be driven not so much by simulation tools that say, “The analysis says don’t do this,” but more by helper applications that say, “Successful designs don’t have the features you are proposing.
There’s a way to go before we implement such co-pilots, but we have built the Cadence Joint Enterprise Data and AI (JedAI) Platform, utilizing a database infrastructure, tools and roadmaps, experience, and insights, to enable it to apply the power of ML and AI to electronics design. We believe it will help designers keep global innovation on track.
Suggested Reading
iSentek’s Cutting-Edge E-compass Sets New Standards with Dynamic Range Breakthrough – EE Times
How Etch is Evolving to Meet the Demands of the AI Era – EE Times
Mastering Portability in AI’s Hardware Revolution – EE Times
Flyking Technology Receives 2023 Outstanding Distributor in Supply Capability Award at IIC Shenzhen – EE Times
Upcoming Virtual Event: PowerUP Energy