Inside the pastel-toned cases of Google’s latest Pixel phones sits the Tensor, Google’s new system-on-a-chip (SoC), and its attempt to compete against Apple’s A-series chips. Like Apple Silicon, Tensor uses custom-designed chips matched to the hardware. In Google’s case, the Tensor includes a new security chip, the Titan M2, and a mobile TPU (Tensor Processing Unit), which is built to run AI processes like Night Sight, and Recorder voice transcription. This looks like the beginning of a trend that may topple the mighty all-purpose chips of Qualcomm and Intel. “Intel, Qualcomm, and other semiconductor suppliers should be worried about the development of purpose-built processors by the likes of Apple and Google. These two events represent a trend toward application-specific processor development and are likely to continue with other, larger technology players,” Harry Pascarella, vice president of IoT analyst and strategy company Harbor Research, told Lifewire via email.
Custom Silicon
The advantages of custom chips are clear. Just compare Apple’s M1-based MacBooks to the previous Intel versions. From the outside, the Intel and M1 MacBooks Air are identical, but the all-Apple model is way faster, has a battery that can run for days on one charge, and runs so cool that it doesn’t need a fan. Partly this is down to Apple’s M1 chip design, which is essentially a continuation of the A-series iPhone chips. These evolved in an environment where extreme efficiency was paramount, and it shows. But another part of the equation is that Apple’s chips and software are co-designed to complement one another. Whereas as Intel’s x86 chips have to be general-purpose, like a family saloon (yes, we’re bringing out the car analogies), Apple’s chips, and now Google’s Tensor SoCs, are matched to the software they run. They’re like finely tuned sports cars that also get great gas mileage. Or something. This kind of synergy is impossible when you’re running an off-the-shelf operating system (Android or Windows) on general-purpose, off-the-shelf hardware (Intel, AMD, Qualcomm). But does this mean the end for these general-use chips? Not at all. They still shine when it comes to flexibility, for example. Here’s another car analogy, this one stolen from Steve Jobs when describing the difference between PCs and tablets. An Intel x86 chip is a truck, whereas the custom silicon is a sports car (again). “Generic processors will continue to be useful, particularly in PC, laptop, and server applications. However, in IoT, mobile devices, and other devices, purpose-built processors are going to continue to increase in prevalence,” says Pascarella. One twist is that Intel runs on a model similar to Apple’s. It designs the chips, and it builds the chips. Apple’s ARM-based silicon is designed by Apple, but then fabricated by a third-party manufacturer, in this case, Taiwan Semiconductor (TSMC). In the past, this has been to Intel’s advantage because you could only get Intel chips from Intel (although other x86-compatible chips are available). One possible path for Intel is to become a chip fabricator like TSMC, but that would turn it into just another factory, competing on price. Add to that the fact that Intel isn’t actually very good at making chips right now. It is currently a generation behind TSMC and even plans to pay TSMC to make Intel’s chips in 2023.
Android and Tensor
Meanwhile, Google could license Tensor the way it licenses Android. That would remove the advantage that its new Tensor-based Pixel 6 handsets enjoy, but it would also close the gap between the iPhone and all non-Google Android phones. And with Apple cracking down on the kinds of privacy violations that make Google’s core advertising business tick, strengthening Android, in general, makes a lot of sense. For you and me, this is all good news. These chips not only make our computers faster, cooler, and with better battery life, but they also allow features not possible before, like Google’s AI photography enhancements and Apple’s incredible new on-device Live Text.
