Yesterday in my post Intelligent Systems, I wrote about how the imperative for differentiated products, especially at the high end of markets, is pushing both system and semiconductor companies to take a more holistic view of system design. Cadence calls this "Intelligent System Design".

Design Excellence

Of course, the heart of designing electronic systems are the tools and IP required for creating the various parts of the system to be optimal. In semiconductor design, this is often referred to as PPA, for Power, Performance, and Area (which is also a surrogate for cost). This is not new, and is the heart of everything that Cadence does. We call this Design Excellence. Designing the best chips, packages, and boards is the most important thing, everything else is secondary.

Examples of recent advances in our product portfolio are:

• World-class PPA results with Cadence digital flow resulting in production deployment in 17 of the top 20 semiconductor and system companies. Full flow delivers 10-20% better PPA with the best turnaround time. The Genus and Innovus solutions are merged using the iSpatial technology, which allows limited synthesis to take place under the hood of place and route, further improving the unified physical optimization flow.
• The Spectre X Simulator provides an up-to-10X speedup while maintaining the golden accuracy, as well as up to 5X increased capacity. The Spectre family is the most widely used simulators in the industry. While getting the same result faster doesn't directly improve PPA, it allows more simulation to be done to achieve a more optimal result. 
• The Protium X1 Enterprise Prototyping Platform allows embedded software and SoCs to be developed in parallel. More than that, even when prototype or production silicon is available, the prototyping environment is often a better environment for getting to the root cause of any issues that might arise.

System Innovation

The next level up is the system itself. Some parts of this, such as PCB and package design, Cadence has done for years. But, increasingly, it is not enough. While Cadence is not going to suddenly start designing phone cases or fans, modern systems require all the electronics to be analyzed in the context of the environment in which they run. If you drill holes in a camera case, you might improve the thermal properties since air can circulate more easily, but you might make EM interference worse since there is less shielding.

The underlying technology for system analysis is very similar to the underlying technology for things like on-chip IR analysis in a product like the Voltus IC Power Integrity Solution, handling extremely large sparse matrices in a computationally efficient manner. This is based on finite element analysis, a technique that has been around since the 1940s where physical structures are broken down into a mesh of tiny elements that are small enough to be analyzed separately and then combined.

The first product Cadence announced in the system analysis space was the Clarity 3D Solver. See my post Bringing Clarity to System Analysis for details when we announced it. The basic techniques are broadly applicable.

The other big aspect of system innovation is software. The Protium and Palladium platforms allow for software-driven SoC verification, as well as debugging the software. The Portable Stimulus Standard (PSS) and Perspec System Verifier allow verification to be pushed up a level to booting operating systems like Android and Green Hills Software's Integrity.

Pervasive Intelligence

One of the biggest developments of the last decade has been the blossoming of machine learning techniques. The basic ideas have been around for decades, but the big increase in available compute power and some major breakthroughs in how to train neural networks has made these approaches much more broadly applicable. Artificial intelligence (AI) is being added to many products such as smartphones and security cameras, and creating new product segments ("Alexa play Taylor Swift").

Increasingly, a lot of this AI is being moved out of the cloud to the edge. This is driven by a number of forces. One is people are more concerned with privacy and increasingly unhappy about uploading everything to the cloud. But, especially with smartphones and ADAS, there is a lot of aggregate compute power at the edge. But moving inference to the edge typically means special AI processors and often special chips tailored to just the workload required (voice recognition, traffic sign recognition, and so on).

A lot of designing an AI chip is the same as designing any other chip. A DDR5 interface doesn't care what you do with the data. But Tensilica DSPs in particular supply a range of specialized offload processors for functions like processing of audio, video, radar, and general neural networks. These processors are used under the hood to offload the main power-hungry microprocessor since they are hundreds or even thousands of times more efficient. We call this machine learning enablement.

But these trends don't just affect Cadence customers. They affect tools and design flows. Inside the tools, machine learning techniques can be used to improve the algorithms and get better PPA. This is invisible to the tool user, other than better results, just like any other improvement in the algorithmic internals. We call this machine learning inside.

Design flows can be improved by machine learning, too. Historically, each run of a tool is starting from scratch with no access to the history of previous runs of the design, nor of similar designs. The tool is run, engineers inspect the result, they adjust some parameters, and the tool is run again. However, a lot of that iteration can be improved by retaining a lot more information from previous runs and automating some of the parameter adjustment. The ultimate goal is "no human in the loop", where a job is launched and after burning up a lot of compute resources produces the result. Obviously, the intermediate and more realizable goal, is fewer humans in the loop, requiring less intervention by skilled designers. We call this machine learning outside.

Intelligent System Design

Putting it all together, we arrive at Intelligent System Design. We started from a core of Design Excellence, our traditional portfolio of EDA tools and IP. In fact, we started lower still, from competence in computation on huge amounts of data. We’ve expanded out to innovate in the area of system design and analysis—System Innovation. Now we are taking Pervasive Intelligence and using it to enable our customers to incorporate machine learning into their products, as well as putting it inside our products and producing optimized design flows and methodologies.

Learn more about Cadence's Intelligent System Design strategy.

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.