Preparing For Bigger Changes Ahead

The semiconductor industry has undergone a fundamental shift over the past year, and it’s one that will redefine chipmaking over the next decade or more.

While the focus is still on building the fastest, lowest-power devices, whether that’s by shrinking features or packaging them into blazing-fast 2.5D or fan-out configurations, these devices are being customized for specific use cases much more than in the past.

At the introduction of Moore’s Law, and even in the foundation development years before that, the big challenge was to put electronic devices together and make sure they would work. CPU ASICs were the stars, and everything else was of lesser importance. Memory was a commodity. I/O was a wire. And that basically summed up the von Neumann architecture. You could use it to build a faster computer, but you couldn’t do much else with it.

Fast forward to 2017 and the tech world has changed rather significantly. And while it’s true that these technologies have been under development on and off for the past three decades, over the past year markets really began redefining themselves and new technologies began unfolding. In this brave new world, CPUs have much more limited roles. They are augmented by accelerators of all types, whether they’re FPGAs, eFPGAs, MCUs, GPUs or DSPs. New memories continue to hit the market, including MRAM, ReRAM, STT-RAM, and phase-change varieties such as XPoint. Even DRAM is available in multiple configurations, with the HBM3 spec now locked and DDR5 under development. Plus, there are new materials under development for interconnects and substrates, new lithography technologies rolling out, and new device structures such as gate-all-around FETs on the horizon.

All of these pieces are essential for the next phase of development, which include narrowly defined end markets and brand new areas of growth. This is no longer just about building a faster computer or shrinking what used to be a supercomputer down into a smartphone with a battery that lasts all day. That work will continue, but the focus is more about optimizing this technology for specific purposes. In effect, it’s no longer just about the technology. It’s what you can do with that technology, and that’s a very big deal.

There are multiple implications that go along with this shift. First, there are entirely new markets with unique technology challenges that have never been commercially viable in the past, such as automotive electrification, AI, machine learning, industrial IoT. One of the big challenges for 2018 will be developing systems to process, store and access large amounts of data more quickly, and that will involve more than just creating faster chips. It will require completely new architectures, from placing memory closer to optimized processing elements to possibly putting some processing capability inside of the memory itself. These architectures will have to be both customized and flexible, because many of these markets are so new that standards and different approaches are still in flux.

Second, while software increasingly defines system architectures, functionality increasingly is being added into hardware. Hardware is faster, it uses less power, and it’s more secure. The pendulum has swung firmly back to the hardware side, with software now taking advantage of customized, increasingly heterogeneous mixes of hardware components rather than trying to leverage generic hardware with customized software. This is a big shift by itself, and it will have a profound effect on system architectures.

Third, and perhaps most significant, chips increasingly define larger systems. The result is that chip technology is moving much closer to the end customer, or in the case of companies like Apple, Samsung, Google, Amazon, Facebook and Microsoft, chips are being designed by systems companies themselves. Whether that remains in-house, or whether larger parts of that are outsourced remains to be seen. But what has changed is that chip design is increasingly application-specific, and those applications increasingly are defined by individual companies rather than by broad markets.

This bodes well for semiconductors in 2018 and far beyond that. There will be multiple waves of technology required to enable some of the technologies that will be created over the next decade, and new markets that will open up as some of these technologies mature. And from a system-level perspective, system-level design is now very real, but the definition of the system is very different than it was a couple years ago.