Chipmakers turn to the next little thing

As it does every year, the recent Mobile World Congress included some big phone announcements. The Samsung Galaxy S7 and LG G5 both look like strong updates, and they are among a long list of new phones that will be using brawny processors such as Samsung's Exynos 8890 and Qualcomm's Snapdragon 820.

But what struck me about the news coming out of Barcelona is how rapidly the show has expanded beyond phones to emerging areas such as the Internet of Things, as well as the wireless infrastructure that will be need to support billions of additional devices of all shapes and sizes. These devices need different kinds of processors--ones that trade raw performance for the smallest size, lowest cost, longest battery life, and integration of features such as connectivity, sensors and security.

So it's no surprise that we are starting to see lots more action around these kinds of chips including numerous announcements at Mobile World Congress as well as Embedded World, a conference that took place simultaneously in Nuremberg, Germany.

Perhaps the biggest little announcement was ARM's latest addition to the Cortex-A series processors used in billions of phones and other devices. The new Cortex-A32 sits between the tiny Cortex-M series microcontrollers used in devices running an embedded operating system and the more powerful Cortex-A35 and A53 processors used in entry-level smartphones. It is based on the same ARMv8 architecture as the rest of these recent A series designs, but it is 32-bit only. ARM's Ian Smythe, director of CPU marketing, said in an interview that there is a large ecosystem of 32-bit embedded devices where "every cent and every millimeter matters."

By dropping 64-bit support, ARM was able to create the smallest and most-efficient Cortex-A series core ever. On a 28nm manufacturing process, a single Cortex-A32 core takes up just 0.25 square millimeters and runs on just a few milliwatts at 100MHz, though it can also scale to a 1GHz quad-core that uses less than 75 milliwatts. (To put that in perspective, a high-end smartphone processor uses anywhere from around 100 milliwatts to several watts when the fastest cores are running simultaneously under heavy workloads.) The result is a chip that is much more efficient than the low-power Cortex-A5 and A7 but matches the performance of the 64-bit Cortex-A35 because it supports the same instructions.

The Cortex-A32 can be combined with ARM's Mali-470 graphics (and Mali-V550 video and display processors) in high-end embedded systems. ARM is targeting what it calls "rich embedded devices" such as wearables, and smart home and industrial IoT devices, though it has yet to announce any licensees or products. ARM also announced the Cortex-R8, which it said will provide twice the performance of its current real-time processor for applications that require high throughput and low latency such as LTE-Advanced and 5G cellular modems, and enterprise storage.

Competitor Imagination Technologies did not make any CPU announcements, but it did unveil its latest generation PowerVR XE Series graphics, which is also targeted at rich embedded devices as well as low-end phones. The PowerVR Series8XE is based on the same Rogue architecture as the Series8XT used in high-end phones, but the sorts of devices in which it will be used don't require gigaflops of gaming performance. So Imagination kept the fillrate the same and scaled back the shader programs until it reached certain targets for area and cost. (Fillrate refers to the raw number of pixels that can be rendered on the screen per second while a shader tells the device how to render things on the screen.) The Series8XE delivers the same fillrate as the current 7XE in a smaller area, and higher performance than competitors in a smaller area, making it well-suited for TVs and set-top boxes, automotive infotainment systems and clusters, and entry-level phones and tablets.

MediaTek is the first Series8XE customer, though Peter McGuinness, Imagination's director of multimedia marketing, said in an interview that there are multiple licensees in automotive, consumer electronics, mobile and enterprise. McGuinness added that the Series8XE's support for 10-bit YUV color makes it an inexpensive way to add HDR (high-dynamic range) to 4K TVs. Automakers also like the OmniShield hardware virtualization, which keeps the infotainment separate from critical car systems. The Series8XE supports the newest APIs including OpenGL ES 3,.2 and its successor, Vulkan 1.0, an open, cross-platform alternative to DirectX 12 and Apple's Metal for squeezing more performance out of the graphics hardware and reducing the load on the CPU.

With the recent acquisition of Freescale, NXP has gained a broader line of ARM- and Power-based processors for embedded applications. At Embedded World, the company announced a new version of one of these, the QorIQ LS1012A processor, that combines a single 64-bit ARM CPU with high-speed Ethernet in a tiny, power-efficient package. The 28nm chip includes an 800MHz Cortex-A53 core; 2.5Gbps Ethernet, PCI-Express, SATA3 and USB 3.0; and hardware IP packet acceleration to reduce the load on the CPU and stretch battery life in a package that measures less than 100 square millimeters and uses around one watt of power.

NXP claims the LS1012A is the smallest and most efficient 64-bit network processor available. This makes it well suited for embedded devices that need both high-speed communications and low power such as home automation and security, IoT gateways, and portable wireless drives.

NXP noted that it also is the first high-performance network processor small enough to fit on a hard drive's printed circuit board converting it to an Ethernet-connected "smart drive" for software-defined storage applications. "You can out Ethernet directly on the hard drive and do all the management in software that takes care of where everything goes," said Jim Bridgwater, a senior product manager at NXP, in an interview. The LS1012A, which is part of a line of 64-bit ARM processors that scales all the way up to an eight-core Cortex-A57 (soon moving to Cortex-A72), will start sampling to key customers in April and will be in mass production in the fourth quarter.

At the extreme low-end, Cypress announced a new addition to its PSoC (programmable system-on-chip) for embedded devices that are upgrading from 8-bit microcontrollers. The PSoC-4S has a 32-bit Cortex-M0 processor, 16- to 32KB of flash memory, 36 general-purpose IO pins and the company's CapSense capacitive sensing controller. ARM has previously referred to the M0 as "specks of dust" because the chips are so small that they can fit inside the dimple on a golf ball. These chips, which sell for $1 or less, can be used in a wide range of embedded applications including automotive, consumer electronics, computer peripherals, medical, and industrial IoT and lighting.

There is clearly a need for capable processors that are smaller and that can last for days or weeks on a single battery. The volumes will be huge--estimates for IoT and wearables are all over the map, but the number if certainly in the billions. However, these chips don't require leading-edge process technology--most are likely to remain on 28nm for a long time since it is so cost effective--are very small, and sell for only a few dollars. So whether this market will be enough to fill the fabs and offset declines in chips for PCs and phones remains unclear.