Mach42’s ML-Powered Analog Verification Breakthrough

Revolutionizing Analog Design Verification with Mach42's Discovery Platform

Analog design verification has long been a challenging and time-consuming process. Despite the significant advances in the field, one persistent bottleneck has been the need for accurate but fast simulations. For years, engineers have relied on SPICE simulations for design verification, but the massive computational demands and long run times often hinder rapid design exploration. Enter Mach42's Discovery Platform, a game-changing innovation in analog verification, leveraging machine learning (ML) to streamline the process and accelerate design exploration.

Why Analog Design is Hard

Traditional verification for analog designs relies heavily on SPICE simulations. These simulations are incredibly accurate but come at a significant cost: long processing times and the need for vast computational resources. Analog designs involve a multitude of variables — from voltages and currents to temperature variations and RLC components — all of which must be validated across a broad range of conditions. Historically, this has required Monte Carlo (MC) analysis or the more recent Scaled Sigma Sampling (SSS) methods, both of which demand huge resources and long processing times.

While these methods are essential for final design signoff, they pose significant challenges during the early stages of design exploration. Designers are left with limited time to experiment and iterate, especially when project timelines are tight. The need for a faster and more flexible solution has never been more pressing.

Mach42's Discovery Platform: A Breakthrough in Analog Verification

Mach42’s Discovery Platform is a revolutionary ML-based solution that aims to change how analog design verification is approached. Instead of running full SPICE simulations every time, the Discovery Platform builds a surrogate model using machine learning. By training the model on existing simulation data, the platform can quickly produce results that are 90% as accurate as full SPICE runs.

Here’s where it gets interesting: The platform isn’t just about speed. It intelligently learns from the data to focus on regions of the design that show high variance, ensuring that the surrogate model remains highly accurate in critical areas. With active learning and a reconfigurable neural net architecture, the Discovery Platform adapts and improves with every run, allowing designers to quickly explore design options without sacrificing accuracy.

The result? Rapid iterations with good-enough accuracy during the design exploration phase, all while maintaining the option for full SPICE confirmation when it’s time to sign off on the design. This dramatic speed-up is essential for modern design cycles, where speed-to-market is a top priority.

Expanding Beyond Analog: The Potential for Analog-Digital Integration

Mach42’s Discovery Platform isn’t just about speeding up the analog verification process; it also holds the potential to bridge the gap between analog and digital design. One of the most exciting aspects of this platform is its ability to work at the system level, including package and board parasitics. This capability means that analog designs can be verified in the context of the entire system early on, making it easier to identify potential issues related to power, voltage, or signal integrity before they become costly problems.

The Road Ahead: Collaboration with Cadence and the Future of Verilog-A Models

Mach42 is already well on its way to transforming analog verification. They’re collaborating with Cadence on Spectre, and the Discovery Platform is undergoing active evaluations with multiple customers. But the innovation doesn’t stop there.

Mach42 plans to take things a step further by developing Verilog-A models as part of this platform. These models will be essential for AMS (Analog Mixed Signal) designers and could even open the door to new forms of digital verification. The ability to develop RNM (Reduced-Order Models) for hardware accelerators could have a profound impact on the analog-digital divide, enabling comprehensive testing across both domains in ways that were previously impossible.

The Future of Analog Design Verification

Mach42’s Discovery Platform is poised to redefine how analog designs are verified. By combining the speed of ML with the accuracy of traditional SPICE simulations, the platform allows for rapid exploration of design options while retaining the reliability of full simulations when needed. The potential for system-level verification and the development of Verilog-A models sets the stage for a new era of analog-digital integration that could transform the entire semiconductor industry.

As the pressure to reduce design cycles and accelerate time-to-market continues to mount, innovations like the Discovery Platform are more than just useful — they’re essential. Mach42 is paving the way for faster, smarter, and more efficient analog verification, and the industry is watching closely.

Source - Semiwiki