Processor IP Vendors Adapt to IoT
…with smaller CPU and GPU cores, better security and new radios
After years of making their cores bigger and faster, processor-IP vendors are now looking for the next little thing. The seemingly insatiable consumer appetite for higher-performance mobile devices has long driven the industry’s processor-IP roadmap, but as growth in smartphone and tablet shipments slowed, vendors began shifting to leaner CPUs and GPUs for the Internet of Things (IoT) and wearables. These markets have great potential for growth: our forecast, which is more conservative than most, expects total shipments of IoT and wearables to exceed 1.5 billion units in 2020 and to surpass smartphone shipments by 2023, as Figure 1 shows.
CPUs and GPUs Get Small
Whereas ARM counts Cortex-A53 as a “little” core, it’s still too much CPU for IoT and wearable devices. The company’s new Cortex-A35 targets those applications with a smaller CPU that it estimates will consume just two-thirds the power of the A53 at the same clock frequency. The new design uses a simpler single-issue microarchitecture that shaves 2 percent off the A53’s die area, making it the smallest 64-bit CPU on the market. The company targets the A35 at high-end IoT and wearable devices.
For lower cost and power, ARM offers its popular Cortex-M lineup, but it faces increased competition from a new crop of cores vying for IoT applications. The new Andes D10 includes extensions to the company’s base ISA for DSP instructions, making it a contender against Cortex-M4. Synopsys is also taking on the M4 through the enhanced DSP capabilities of its tiny EM-series MCUs, which include dedicated X/Y-variable memories in the EM9D and EM11D.
Imagination’s new Warrior M6250 provides optional MIPS DSP extensions with 32-bit SIMD support. It includes an L1 cache controller and TLB MMU, which Cortex-M cores lack, but these features consume more die area and power than they do in competing products. The M6250’s MMU option supports running a high-level operating system such as embedded Linux. The new M6200 reduces die area by eliminating the integrated cache and MMU, making it well suited to use with an RTOS in low-cost microcontrollers.
Although many IoT markets are just getting started, MCU shipments are already taking off. Shipments of 32-bit MCUs surged by 42 percent compared with 2014, more than double the rate of mobile chipsets. The overall embedded segment—which includes 32-bit MCUs, high-end smartcards, and flash-memory controllers—increased by 30 percent. In our updated forecast, we expect the embedded segment to zoom past the maturing mobile segment in 2016, with IoT devices being a major contributor to this growth, as Figure 2 shows. By the end of 2017, embedded-processor shipments will double compared with 2014 and will comprise roughly 40 percent of the CPU-IP market.
GPU-IP vendors are also changing their focus from high-end graphics cores to smaller low-cost cores for IoT and wearable devices. For example, the new PowerVR G6020 is Imagination’s smallest Series6XE GPU, but it maintains full OpenGL ES 3.0 compatibility. Support for the latest 3D-graphics API makes it suitable for high-end smartwatches and IoT devices with up to 720p displays, such as home security systems, as well as entry-level smartphones.
ARM took a different approach with its low-end Mali-470 GPU, tweaking the Mali-400’s Utgard architecture. The new core is a power-efficient replacement for that 2008-vintage design. Utgard supports only the older OpenGL 2.0 ES, but that standard is adequate for the limited 3D-graphics requirements of most wearables.
Radioing for Help
Despite the diversity of products that comprise the Internet of Things, those products all have one common requirement: a means of connecting to the Internet. We expect most IoT devices will connect wirelessly to local gateways using Bluetooth, Wi-Fi, cellular, or IEEE 802.15.4 variants such as 6LoWPAN, Thread, and ZigBee. To minimize cost and power, chip vendors must integrate the radio in the processor. Such designs require specialized RF design expertise, which most CPU developers lack. To fill this gap, several CPU-IP vendors acquired radio IP and expertise from small RF design houses in the past year.
Last April, ARM purchased two suppliers of Bluetooth IP: Sunrise Micro Devices and Wicentric. Sunrise developed the Cordio BT4 Bluetooth radio as a hard macro optimized for TSMC’s 55nm CMOS process. Wicentric formed a partnership with Sunrise to provide the BT4’s link layer. ARM plans to combine the Cordio hard IP with its Cortex-M synthesizable CPU to build an IoT processor reference design.
Imagination’s Ensigma products comprise modem IP for a variety of wireless standards; the company calls them radio processor units (RPUs). Also in April, the company released the new Ensigma Whisper products, including the C5300 for Bluetooth Smart and the C5400 for 2.4GHz IEEE 802.11n Wi-Fi. The Ensigma C5401 combines both radios and enables concurrent operation. Imagination separately licenses the RF front-end macros, and customers can also license them from third-party vendors.
…embedded segment to zoom past the maturing mobile segment in 2016, with IoT devices being a major contributor to this growth.
Ceva customers must use radio cores from third parties, but the DSP-IP provider worked with several RF design houses to develop a family of Bluetooth and Wi-Fi products for 28nm to 65nm processes. The company combines the hard macros with synthesizable digital basebands and media access controllers (MACs) developed by RivieraWaves, a four-year-old French startup it acquired in 2014. The company’s Sense, Surf, and Stream products cover three different Wi-Fi performance and power levels. The breadth of the Ceva and Imagination lineups far surpasses that of ARM’s single-node Cordio.
Enhanced Security for IoT
IoT growth is raising awareness of security risks in connected devices. The new ARMv8-M ISA adds a host of new features to strengthen security in Cortex-M-based processors. The enhancements integrate IP from the company’s 2015 acquisition of Sansa Security, an Israel-based developer of an IoT security platform that ranges from client devices to the cloud. ARMv8-M retains a 16/32-bit architecture, making it suitable for embedded controllers. It optionally provides TrustZone, which was previously available only in Cortex-A.
Imagination’s OmniShield software and hardware technology surpasses TrustZone by enabling its processor cores that support hardware virtualization to run multiple secure computing environments. These secure environments go beyond MIPS CPUs to include applications running on PowerVR GPUs as well as Ensigma modems. Each isolated domain supports bare-metal operation, an RTOS, or a rich operating system such as Android or Linux, enabling programmers to isolate applications in each domain.
Synopsys released an enhanced ARC EM security package, SecureShield, for the EM4 and EM5D cores. This technology includes a new memory-protection unit (MPU) for implementing trusted execution environments, which can separate secure and nonsecure read, write, and execute functions as well as enable per-region scrambling and encryption for up to 16 memory regions. SecureShield restricts access to the ARC processor’s AHB system-bus interface from other SoC components, and it supports in-line data and instruction decryption along with address scrambling to prevent IP theft.
Until recently, SoC designers developing for IoT and wearable devices had to make do with IP that was intended for smartphones or general-purpose MCUs. These recent developments, however, show that IP vendors are now focusing on these emerging markets, which have unique requirements for performance, power, and cost. The acquisition of several radio-IP vendors by larger IP companies allows SoC designers to get more of their IP from a single source, simplifying their task and ensuring compatibility among the various cores. The next wave of IP cores are likely to offer greater integration and new features to optimize the development of IoT processors.
Mike Demler is a senior analyst at The Linley Group, where he contributes articles to the weekly Microprocessor Report and Mobile Chip Report. Demler is also a co-author of A Guide to Processors for Wearables and IoT and A Guide to CPU Cores and Processor IP. He is the program chairperson for the Linley IoT Conference. Before joining The Linley Group, he gained extensive engineering and marketing experience at Texas Instruments, General Electric, Cadence, and Synopsys. He holds eight U.S. patents and is the author of the textbook High-Speed Analog-to-Digital Conversion. He holds a BSEE degree from the State University of New York at Buffalo, an MSEE from Southern Methodist University, and an MBA from San Jose State University.