AI-based analog optimization
AI-based approaches perform well in solving optimization problems where traditional algorithmic approaches fall short. AI has the ability to automate manual loops in the design process. Like a human designer, AI learns and learns from experiments, integrating learning across each experiment to understand and navigate the solution space. In general terms, this approach is called sample-based optimization.
Model-based approaches such as grid search, i.e. parameter swapping, and random search, i.e. Monte Carlo simulation, have traditionally been used to assist designers during the analog design process. However, these approaches do not scale well. The number of samples required for sufficient solution space coverage scales with design complexity.
There are more efficient general methods, such as Bayesian Optimization, which are widely used in machine learning applications. A Bayesian Optimizer constructs a probabilistic model of the objective function and uses it to select new sample points in the metric space with a high probability of scoring well. Thus, it learns from previous patterns to build a model that helps select future patterns.
An AI-based approach represents an even more focused, intelligently directed way to navigate a large and complex solution space to find sample points that meet the specification. An AI capability can be conceptualized as a model-based optimization system that dynamically learns about the problem it is tasked with solving.
Such an AI approach can use real, multi-corner/multi-testbench simulation to drive complex corner and testbench dependency searches. It can dynamically navigate process vertices to reduce the number of simulations required by connecting all vertices. Through this process, the AI tool learns from its simulation experiences, using a live feedback loop to gravitate toward a solution that meets the specification.
A key advantage of such an AI system is that it does not depend on a specific form of the problem it is optimizing. However, unlike less efficient sample-based methods, it will adapt itself more effectively to the underlying purpose. It doesn't even optimize proxies, but rather runs on actual circuit simulation.
Such a system is possible because the AI system makes informed decisions based on the experiments it runs, which reinforces its internal view of the problem and objectives, enabling rapid coordination.
AI-based analog migration
Macro trends, including the slowdown of Moore's Law, manufacturing capacity constraints, and a challenging geopolitical climate, are driving the need for new capabilities to rapidly move designs between process nodes.
To take advantage of market opportunities and be resilient to supply chain challenges, semiconductor companies must develop an agile supply chain landscape, including moving products from one foundry to another and from one technology node to another. Porting to another. While AI can help speed up and automate circuit optimization in general, it has a distinct advantage during design transitions.
As explained in Figure 1, the analog design transfer process begins with a reference design, specification, schematic, and layout in the specific technology node, and ends with a complete and functional layout in the target node. The challenge of moving an analog circuit from one node to another is different from typical analog circuit design, in that the circuits are based on an earlier version of the design.
This is good news for AI: any design that has been optimized in a context has valuable learnings that are useful, even if the context, such as a technology node, has changed.
Figure 2 Describes an AI-powered, automated design transition process. The first step is to transfer to the target node. Circuit elements and transistors are mapped to equivalent elements in the target node, the specification is adapted to the target, and the design is parameterized with the parameters needed to adjust the circuit to the specification.