The new semiconductor consumers represent a very different demographic. They are less concerned about chip pricing than they are about performance, power, and time-to-results. To understand this phenomenon, let’s look at a few examples:
Apple (new Mac processor)*1
Apple was one of the earliest companies to see the strategic benefit of custom silicon. They brought processor design in house over a dozen years ago. The initial focus was on cell phone processors, and the iPhone made history. Today, they’re building processors for their computer gear. Apple’s first SoC processor for Macs represents a 119mm2 M1 die with 16 billion transistors. That’s not a typo. 16 billion.
Google (accelerate AI workloads)*2
In 2016 Google introduced the first tensor processor unit, or TPU. The device aimed to accelerate AI workloads using its TensorFlow software. Both training and inference applications were contemplated. The current TPU (v3) has a matrix architecture and can deliver up to 99% scalability in huge, 1,024-chip configurations. It handles some of the largest AI workloads in the world and consumes 450 watts. TPU v4 is expected to be available later this year and is rumored to be built in 5nm or 3nm technology. Through this massive investment, Google is committed to global leadership in AI.
Amazon Web Services*3
AWS is focused on AI in its datacenters and at the edge. Their 7nm 64-bit Graviton2 custom processor contains 30 billion transistors and occupies about 350mm2. The device is a key part of the company’s Elastic Compute Cloud (Amazon EC2). The investments don’t stop here though. Other chips in development include:
- Graviton3 – 5nm or 3nm
- Trainium AI training processor – one trillion calculations/sec; 30% higher throughput, 45% lower cost than GPUs
- Inferentia Arm®-based AI inference processor; 25% higher throughput, 30% lower cost than GPUs
This small summary of what’s going on at the new semiconductor consumers tells a take-no-prisoners story. Aggressive, no-compromise projects like the ones summarized here create a vibrant outlook for both semiconductors and EDA. And finally, my non-technical friends and family understand what I do.