The Next Phase Of Computing

Apple’s new M1 chip offers a glimpse of what’s ahead, and not just from Apple. Being able to get 18 to 20 hours of battery life from a laptop computer moves the ball much farther down the field in semiconductor design.

All of this is entirely dependent on the applications, of course. But what’s important here is how much battery life and performance can be gained by designing hardware specificially in conjunction with the software, rather than each being designed separately based upon some general-purpose connection scheme, such as a general-purpose chip, running a general-purpose OS, using general-purpose APIs.

The fact that the M1 chip is based on a 5nm process is good marketing, but that by itself does little for the overall device performance or energy efficiency. Just having more transistors packed on a die doesn’t mean much without incredibly fast interconnects between the ultra-dense processing elements and memories, or without an underlying power delivery network capable of getting enough power to all of those processing elements at the same time.

That Apple started out on the low end of its product line with the most expensive process technology is an indication it wants to fine-tune the system in the field for various applications before turning up the heat — literally and figuratively — on performance. In most cases, the most advanced technology goes into the highest-priced, highest-performing device, whether that’s a computer or a car, because the developer wants to recoup its investment as quickly as possible.

While the M1 chip includes a CPU, GPU and NPU, the interesting part will be what happens with customized acceleration for applications such as image and video processing. Apple develops all of this internally, so it has the ability to fine-tune just about everything.

But the company is hardly alone here. In the future, performance and power specs will become much harder to decipher because they will be tied increasingly to specific use cases. There are plenty of such use cases, and so far there are no clear leaders in the markets they will serve, in part because these markets are so new and in part because there has never been an option for this level of customization. The possibilities and number of options is growing exponentially.

Intel, AMD and Samsung all are heading in this direction. So is Huawei, based on chips from HiSilicon. Devices that do some level of computing — and that list is expanding, with rapidly blurring distinctions about what’s a computer and what isn’t due to the emphasis on smart everything — will need to fit into an acceptable power envelope. In the future, that also will include an energy envelope, slimming down processing to only what is required to run at a particular clock frequency, and doing that as efficiently as possible.

Future generations of devices will maximize throughput and access to memory, while optimizing compute cycles for the task at hand. In the future, much of this will be done dynamically as loads and algorithms shift, and as new IP is developed to take on some of these programming challenges.

Put in perspective, this represents a fundamental shift in design across a wide range of applications, which is why the entire tech industry is scrambling for more talent these days. The Apple M1 is a high profile example, but there is much more to come.