Cadence's goal is to empower engineers at semiconductor and systems companies to create innovative, intelligent, and highly differentiated electronic products that transform the way people live, work, and play. One big change is for system companies to design their own chips to enable substantial differentiation. In fact, system and software companies designing their own chips is, although still small, the fastest growing part of the whole semiconductor industry, with a CAGR of 70% over the last five years.

Traditional semiconductor companies, both fabless and IDMs, also face similar challenges to deliver complete systems, software, reference designs, and more. Increasingly, they are adding intelligence to their SoCs to enable their customers to build products that are of more value to their eventual users.

For everyone, the challenge is to deliver new differentiated product faster, more efficiently, despite the growing complexity of both the silicon and the software, and new challenges such 3D packaging and integrated radio antennas.

Adding Intelligence

Mobile and consumer products led the way for this holistic system design optimization. It is notable that the top three mobile companies design their own application processors. See my post Samsung Galaxy S9's Application Processor for a detailed look at one example. More recently, the automotive industry has started to follow. See my post HOT CHIPS: The Tesla Full Self-Driving Computer for the recently revealed details of the FSD chip that they have been working on for the last couple of years. Other segments will follow.

Areas where increasing intelligence is being added include:

• Mobile: Facial recognition, voice recognition, improved photographs, classification of photographs, and more.
• Cloud: Especially for training neural networks.
• IoT/industrial: Smart security cameras, smart thermostats, visual recognition, anticipating failures in jet engines, and so on.
• Automotive: Automated driver assistance systems (ADAS) and future autonomous driving, improving safety.
• Aerospace and defense: Increased automation of unstable airframes, increased safety of commercial airframes, intelligent drones.
• Health: Proactive monitoring for issues, better drug delivery systems...resulting in healthier individuals.

Every system company faces a dilemma: use the same standard products as all their competitors have access to, or design their own SoC optimized for just the purpose they require. The low end of the market can make use of lowest-common-denominator silicon since they are competing primarily on good-enough functionality and other attributes like price and distribution, but the high end really has no choice but to do their own design including chips in leading-edge processes. If you don't need a leading-edge process, then a good-enough standard product in a leading-edge process will outperform a custom design in a mature node.

There is another issue for many, perhaps most, system companies, which is that their future growth is going to come from adding smarts to their systems, making them intelligent systems. This is most obviously true in automotive, with the move to ADAS and eventually autonomous driving. But it is true in many markets where efficiency and privacy are pushing more and more to the edge device: smartphones, security cameras, intelligent assistants like Alexa. Even something as un-smart sounding as a high-end microprocessor is often using neural network technology in the branch prediction. 

Design Flow

The design flow for intelligent systems might start with an environment like MathWorks for signal processing, or perhaps start in something like Caffé or Tensorflow for anything learning related. If the design is not particularly high performance then it can simply be implemented in software and run in a data center. Even in a data center, higher performance requires accelerators, typically implemented in FPGAs or on GPUs, although companies like Amazon are designing their own data center chips in their Annapurna subsidiary (watch for an upcoming post on their HOT CHIPS presentation on AWS Nitro). On edge devices, power constraints mean that the only way to get high performance is to design in a leading-edge process and implement the algorithms in special-purpose processors and, in some cases, add special-purpose hardware, too.

The challenge for many system companies is that they have competence at the system level in whatever domain they operate. That's pretty much a given for serving the market that they are in. What they lack is knowledge of design flows and access to IP required to get from that high-level model captured in TensorFlow, MathWorks, or even just C++, and get that into an efficient implementation. Alongside, there is a need to get the rest of the system running on the hardware. More and more systems are self contained with communication to the outside world by some form of radio, be it 5G, Bluetooth, WiFi, or something else. Radios are the most challenging components to design, since everything depends on everything else. It is not just the shape of the antenna that is critical, but also the shape of every connector, package pin, and PCB trace. Even the parts of the design not directly involved in the RF affect the parts that are, so a holistic approach to design and analysis of the whole system, at the system level, is required.

Cadence cannot design your system for you—only you understand your needs well enough to do that. But we can be your expert partner on how to take your ideas and deliver them as an intelligent electronic system. We have design tools, the key semiconductor IP, the design flows, and decades of experience. And we have more capabilities coming.

We call this Intelligent System Design.

Tomorrow

Tomorrow, a deeper dive into Intelligent System Design.

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