Power Shifts In Digital Chip Space

By Bhanu Kapoor
The power issue has been quite disrupting in the digital semiconductor space. The processor architecture shifted to parallel processing with the “power wall” stopping the frequency scaling that the industry had conveniently used in the last few decades.

The power issue also is causing semiconductor process technology to change in ways other than simply scaling from one generation to the next. First, it was the shift to HKMG technology to address scaling and leakage issues in the process technology. Now we’re seeing a shift to the FinFET process technology (also known as the 3D process technology) by Intel, the largest semiconductor company in the world. The digital semiconductor market dynamics have largely been unchanged so far, but they are likely to see seismic shifts in the future depending upon if and when the foundries catch up to Intel’s process technology lead.

All of the key digital semiconductor market segments rely upon the leading-edge process technology for power, performance, or both. The first one is the market for processors (application and baseband) that go into mobile devices (smartphones, tablets, and laptops) and is probably the most important. The desktop PC need likely will get served mostly by one of these devices with a larger display. The processor going into servers should continue to grow rapidly and become a very large segment itself. The world is just beginning to see the growth of data centers. And the processors going into devices that move the data between mobile devices and servers, i.e., routers and base stations, is another important segment. These three segments should account for the majority of the digital semiconductor market (currently digital semiconductors account for roughly $250B of the $300B semiconductor market) in the near future. The mobile segment likely will get further augmented with wearable devices such as watches, ear buds, and eyeglasses. All of these markets depend upon leading-edge process technology to bring application advances to the market.

The key players in the mobile device processor market include QualComm, Apple (for its own end products), Samsung (for its own end products), Broadcom, Intel, Marvell, Nvidia, and MediaTek. If you look at innovations in the power aspects of digital semiconductors, it seems only Intel is pushing the power frontiers. From the process technology standpoint, Intel will be bringing smartphone processors manufactured using a mature 22nm 3D process technology later this year, whereas the rest of the players will rely on a mature 32/28 nm planar process technology at that time. Intel’s experiments show that a 22nm 3D process technology uses 50% less active power when compared to its 32nm planar process technology and operating at the same performance level. The process technology lead will also give Intel significant power and cost benefits in efforts to integrate a 4G baseband processor with the application processor.

There is an architectural angle to this, as well. Intel has spent a lot of time improving its x86 architecture to catch up with the inherently low power nature of a RISC processor such as those from ARM, which the rest of the players use to build application processors. There is an area where ARM processors are deficient: multithreading. With the advent of multicore processors in the mobile device, simultaneously multithreading is likely to bring 10% to 20% performance gains at similar power levels. The power gains can be equally or more significant.

There are innovations at the package level, also. The integration of VRM in the same package as the processor saves power at the system-level. Intel has done that with the Haswell processor. This is an area where the rest of the industry can catch up, but it is not as simple as dropping the VRM in the same package.

Overall, the digital semiconductor market dynamics have largely been unchanged so far. One key exception is Texas Instruments, which is moving away from the digital space to fully focus on analog semiconductors where the leading-edge process technology isn’t required. There also have been some consolidations among the IP suppliers. Some fabless companies have started to sell IPs, as well, but there have not been many changes for most of the fabless semiconductor companies. Still, given the process technology lead that Intel has—the company is currently a full generation ahead—it remains to be seen if and when the foundries will be able to catch up with Intel. The amount of delay will determine how big a shift the digital semiconductor market will see.

—Bhanu Kapoor is the president of Mimasic, a low-power consultancy.