ARM unveiled a tiny Cortex-A32 processor design based on a 32-bit-only ARMv8 architecture, and offering higher performance per mW than the Cortex-A7.

ARM bills its new Cortex-A32 processor design as the “smallest, lowest power and most efficient ARMv8-A” to date. If that sounds familiar, it’s because it’s essentially the same claim made for the closely related Cortex-A35 announced back in November. The Cortex-A32 bests the Cortex-A35 on all those counts, providing 10 percent higher efficiency (performance per mW) than the -A35, 25 percent more than the Cortex-A7, and 30 percent more than the Cortex-A5, claims ARM. It’s also 13 percent smaller than the -A35.

Cortex-A32 (left) compared to Cortex-A35
(click images to enlarge)

The Cortex-A32 is the first ARMv8 processor limited to 32-bit operation instead of also supporting 64-bit mode, so it can’t achieve the sheer performance of the 64- and 32-bit ready -A35. However, with the help of an 8-stage, in-order pipeline and support for the ARMv8-A AArch32 execution state, the Cortex-A32 can match the speed of the -A35 in 32-bit mode, according to ARM.

Even at 32-bit, ARMv8-A brings performance improvements over ARMv7 chips in areas such as integer, floating point, media, and especially in streaming and cryptography, says ARM. Specifically, the design is optimized for the A32/T32 instruction sets in ARMv8-A’s AArch32 execution state, adding more than 100 new instructions. The Cortex-A32 supports integration with TrustZone CryptoCell-700 series products “to enable enhanced cryptographic hardware acceleration and advanced root of trust,” says ARM.

While the Cortex-A35 is targeting shipments in smartphones starting in late 2016, the Cortex-A32 is aimed at embedded products such as consumer, wearable, and IoT applications. The Cortex-A32 supports Linux (including Snappy Ubuntu Core), as well as Android, Brillo, and Windows IoT. It also supports the Linaro (Linux and Android) toolchain, plus a range of RTOSes including QNX.

Cortex-A32 OS and software compatibility chart
(click image to enlarge)

The Cortex-A32 offers an ARMv8 upgrade path for 32-bit embedded applications, according to an ARM blog announcement. The blog entry adds: “We can expect a significant number of embedded devices to remain 32-bit for the foreseeable future.”


Cortex-A32 comparison chart (left) and performance comparison with Cortex-A5
(click images to enlarge)

Even at its fastest, richest configuration, with four 1GHz cores, 32KB I- and D-caches, 1MB L2 cache, plus NEON and crypto engines, a Cortex-A32 system-on-chip will run at less than 75mW per core, or 300mW total, claims ARM. A similar Cortex-A32 configuration would run at 90mW per core.

In its most minimalist configuration — 100MHz, single-core with 8KB L1 caches — the -A32 is said to run at less than 4mW. As a single-core processor, the Cortex-A32 has a footprint of less than a quarter of a square millimeter, compared to a minimum for a Cortex-A35 processor of 0.4mm squared.

Cortex-A32 quad-core configuration with all the extras (left) and minimalist single-core configuration
(click image to enlarge)

All of the above figures are based on 28nm fabrication. Like the Cortex-A35, A32-based chips will also support 16nm fabrication.

Other Cortex-A32 benefits over the -A7 and -A5 include improved power management logic, a more efficient instruction fetch, and L1 and L2 caches that are claimed to offer 3.75 times better memory streaming performance. The design also has smaller NEON and floating point pipelines with better VFPv4 floating point and DSP performance, claims ARM. The chips will be compatible with ARM Mali GPUs, ARM CoreLink system IP, and ARM CoreSight debug and trace technology

“ARM offers an unrivaled portfolio of processors that power billions of extremely efficient embedded devices,” stated James McNiven, general manager, ARM CPU group. “The Cortex-A32 processor, enabled with secure ARM TrustZone technology, builds on the trail blazed by the Cortex-A5 and Cortex-A7 processors in embedded applications such as single-board computing, IoT edge nodes and wearables. It brings greater performance, efficiency and other benefits of the ARMv8-A architecture for ARM’s silicon partners to innovate on for richer, more secure embedded systems.”

“IoT nodes have become increasingly diverse, with the more sophisticated nodes often requiring a rich OS,” stated Maarten Ectors, vice president, IoT, Canonical. “Combining Snappy Ubuntu Core and the highly-efficient processing and scalability of Cortex-A32 will enable developers to truly push the boundaries of edge devices for IoT.”

 
Further information

No timetable was announced for the arrival of the first Cortex-A32 processors. More information may be found in ARM’s Cortex-A32 press release and blog announcement, as well as its Cortex-A32 product page.