Never miss a story from Corporate and Culture. Subscribe for in-depth analysis and articles.
A core part of what we do at Cadence comes from an inescapable truth: designing and fabricating a silicon chip is an increasingly complex, time-consuming, and therefore expensive process.
At every step of that process there are decisions to be made. How much time or money should be invested at this specific juncture to ensure that when the process is complete, that chip performs as efficiently as possible and that costly mistakes are avoided?
And when it comes to realizing a design that works first time, accurate electromagnetic (EM) simulation is a critical step. But it’s also one that takes the most time – that’s why we created Cadence Clarity 3D Solver software to offer up to 10x faster performance than legacy 3D field solvers, ensuring that complex EM challenges can be tackled at pace.
Using massively distributed cloud technology, parallelization, code optimization and revolutionary mesh generation software, we’ve made the process of designing and running electromagnetic simulations faster than ever before – enabling any potentially devastating EM interference issues to be mitigated against early in the design process.
But as it turns out, removing all the bottlenecks we could from the system itself isn’t enough. Our customers were able to run individual simulations incredibly quickly, but that didn’t inherently mean they were obtaining a more optimized design at the end of the process.
The problem? The process still relies heavily on human intuition and involvement. The result of each simulation, and the steps then taken to identify and adjust parameters, requires a skilled human operator.
This ultimately limits the number of simulations that can be run. Let’s imagine a design simulation that has 10 controllable parameters, and that each of these parameters has 10 possible values. To optimize this design until it cannot be improved further, we need to simulate all possible scenarios and then identify the best-performing combination of values from each parameter.
But 1010 simulations is 10 billion simulations.
Of course, this would take decades to complete. So instead, an engineer will choose values they believe to be close to optimal, then they’ll run a few simulations. They will then look at the results, do some analysis, adjust the parameter values, and run more simulations.
Inevitably, after a number of iterations, they will reach a decision point – continue to optimize further, with all the time and cost that this entails, or declare the design optimized and move forward with production. Undoubtedly, the end product will work well – though perhaps not quite as well as it could have done given infinite time and capacity to run more simulations.
This is the problem we set out to solve with Cadence Optimality Intelligent System Explorer. Could we ultimately enable our customers to achieve a truly optimal design by replacing human involvement within a conventional optimization workflow with an artificial intelligence (AI) algorithm capable of in-design optimization?
It turns out, after several years of hard work from our engineers, that yes, we can. Using reinforcement learning – as used in the Cadence Joint Data and Analytics (JedAI) Platform – we can very quickly build a machine learning (ML) model for what constitutes optimal design of a system of any size or complexity. Reinforcement learning is desirable because it doesn’t require very intensive training. With every iteration, Optimality has more data samples it uses to further refine and improve the accuracy of its algorithm.
The result is a better design, achieved more quickly, more efficiently and with less risk than anything a human operator might have declared to be optimized in the past. With Optimality capable of faster decision-making over multiple concurrent simulations, we’re seeing customers complete their designs on average 10x faster than traditional manual methods, with up to 100x speedup in certain situations and beautiful results typically within 20-50 iterations. This dramatically shorter time window means time to market is reduced considerably.
In one of our tests, we put human designers up against Optimality AI in optimizing a complex PCB design. Optimality was able to arrive at a design that performed significantly better than the human design in only 34 iterations.
Vitally, while human-led simulations could only be completed one or perhaps two at a time, Optimality is capable of running, analyzing and rerunning massively parallelized simulations with little to no pause in between – something that might have required rooms full of engineers in the past.
Optimality is already fully integrated with Clarity 3D Solver and Sigrity X, our high-speed signal and power integrity (SI/PI) platform. By offloading system-level SI and PI simulation and analysis to Optimality, designers can arrive at an optimized design far more quickly – reduce the need for design respins and accelerating overall time-to-market.
Having seen the incredible success our customers are having with Clarity and Sigrity X in Optimality, we’re now working hard to integrate other Cadence products – starting with our Celsius Thermal Solver product, the industry’s first complete electrical-thermal co-simulation solution for the full hierarchy of electronic systems from ICs to physical enclosures.
Again, this is about enabling design teams to hand off optimization of their designs to Optimality AI algorithms, resulting in faster detection and mitigation of thermal issues early in the design process and significantly reducing the time it takes to reach a truly optimized design iteration.
In short order, we’ll also integrate our Fidelity CFD computational fluid dynamics product into Optimality, enabling massively parallel simulations of real-world objects to very quickly arrive at an optimum design with almost no human involvement.
Bringing Fidelity CFD to Optimality will be of huge value to the automotive industry in particular – where there is near-infinite potential to fine-tune vehicles for any number of parameters, from external aerodynamics to internal comfort and powertrain efficiency.
We’re also exploring how to extend Optimality’s technology from the post-layout to the pre-layout stage, giving designers a powerful AI sidekick throughout the early design stages, from concept upwards. This is going to make an even larger impact to your design flow.
This is just the beginning of Optimality’s journey, but in the few months since its launch we’ve seen so many demonstrations of its power to change the way we perform system analysis and the efficiencies that can be achieved.
Which means our original vision – to create a tool that ensures anything from a silicon chip to a jumbo jet performs as well as it possibly can – is only a few iterations away.
Learn more about Cadence Optimality Intelligent System Explorer and how you can achieve a far more optimized design using AI-driven In-design optimization technology.