A trend sweeping through chip design is changing the traditional markets for CPUs and GPUs. Along the way, it is accelerating the number of design starts for artificial intelligence (AI) and building chips known as custom accelerators.
The custom accelerator trend is driven by system companies becoming semiconductor companies not relying on other companies to build their chips. They do most of the chip design in house to control the ecosystem, roadmaps, and time to market, adopting a different perspective. A good example is Apple. It’s designing its own iPhone and Mac chips now and they are not the end product. Their end product is a system that includes peripherals and software.
In today’s landscape, both semiconductor and system companies are racing to make announcements about their own custom accelerators in a vibrant and growing market segment. A custom accelerator is an ASIC, or a chip dedicated to a specific function. It is not a generic CPU or GPU. Nor is it an application serving a broad market. Instead, it is a well-defined, custom-designed accelerator for a specific AI or machine learning (ML) function.
A company uses AI and ML to design its chips and they are not designed by hardware engineers. Because everything starts with a software driver, software defines the product. This new type of user is not taking the derivative of a whole design for targeted chips to address a specific market or application. In this design flow, hardware design is started only after the system specification is defined.
The number of companies making AI or ML chips are generating requirements for loads of verification of big systems because these chips are large and need to run massive workloads and software. The growth is in many areas, not just one. Add hyperscalers as another category not relying on traditional semiconductor companies. All are making their own chips where the size of the workload they run are perfect fits for an emulation and prototyping platform.
