Unleashing the Potential of AI in Simulation

The current technological landscape is saturated with the omnipresence of Artificial Intelligence (AI), Machine Learning (ML), and Neural Networks, raising pertinent questions about the authenticity of their applications. A recent insightful conversation with a client has sparked a profound exploration into the role of AI in simulation, prompting a critical examination of its tangible benefits in contrast to mere hype and vaporware.

The integration of AI into simulation has emerged as a significant advantage across diverse domains. Many simulation products employ AI to mitigate the prolonged duration of simulations, posing a challenge for engineers. The fundamental question surfaces: Is AI indispensable for envisioning additional use-cases when simulations can rapidly execute a full System-on-Chip (SoC) in under three minutes of wall clock time? The swift execution of all required scenarios within this timeframe often surpasses the results generated by AI or Neural networks. Furthermore, simulations provide tangible reassurance by demonstrating thorough exploration of scenarios, as opposed to mere projections, instilling heightened confidence and accuracy.

A recent application of VisualSim Architect from Mirabilis Design serves as a compelling example of the transformative potential of AI in simulation. Crafting a SoC model featuring 128 RISC-V cores and a crossbar, the team ensured cache coherency across eight domains with a 64KB L2 cache at each domain. The evaluation focused on the increase in bandwidth consumption as the cache size expanded and workloads were adjusted across all cores. Through the strategic distribution of the application across domains using a well-designed task graph, the team successfully minimized latency for large matrix processing. Notably, simulation time exhibited a manageable linear rate increase, eschewing an exponential surge.

The crux of achieving such efficiency lies in the thoughtful design of the Discrete-event simulator, event calendar, and the algorithm managing the next event. AI assumes a pivotal role, reminiscent of branch prediction in pipeline loops. Genetic algorithms and AI significantly contribute to predicting the next event, with rapid learning leading to the swift reduction of mispredictions.

Simulation tools, exemplified by VisualSim Architect, adeptly capitalize on these principles to deliver exceptionally fast simulations. Full SoC simulations, encompassing multiple concurrent Task graphs, can be completed in less than 10 minutes. The strategic application of AI in simulation emerges as a powerful ally, offering substantial gains when employed judiciously, heralding a new era of efficiency and confidence in simulation technology.