Embedded computing startup Efficient has unveiled its first standalone processor, the Electron E1 \u2014 which, its creators claim, offers “up to 100\u00d7” the energy efficiency of Arm Cortex-M33 and Cortex-M85 chips.<\/p>\n
“Based on a decade\u2019s worth of research at Carnegie Mellon University, we built our Fabric architecture from the ground up to deliver radically better energy efficiency for general-purpose computing applications,” claims Efficient’s Adam Kaufman of the company’s launch. “The Electron E1 processor is built on this spatial dataflow architecture, which executes general-purpose code while eliminating the need for costly step-by-step computation.\u200d The result is up to 100\u00d7 improvement in energy efficiency over traditional low-power CPUs, enabling intelligent applications at the edge with years-long lifespans in environments where power and maintenance are limited.”<\/p>\n
Efficient claims its Electron E1 chip, built atop the in-house Fabric architecture, can deliver up to a hundredfold efficiency gains for embedded systems. (\ud83d\udcf9: Efficient)<\/p>\n
The first standalone implementation of Efficient’s Fabric architecture, the Electron E1 is claimed to deliver 21.6 giga-operations per second (GOPS) of compute at 200MHz when in high-voltage mode and 5.4 GOPS at 50MHz in low voltage mode. The part includes 128kB of ultra-low-power cache memory, split into 8kB banks, 3MB of static RAM (SRAM), and 4MB of magnetoresistive memory (MRAM) as non-volatile storage, in place of higher-power flash memory.<\/p>\n
The processor’s scalar core, installed alongside the proprietary Fabric hardware, is designed around the free and open source RISC-V architecture, implementing the RV32IAC set plus the Zmmul extension for more efficient multiplication operations. The chip runs on a 1.8VDC supply voltage with selectable internal operating voltage between 0.55\u20130.8VDC. There are a total of 72 general-purpose input\/output (GPIO) pins, six quad-SPI, six UART, and six I2C buses, plus an on-board real-time clock (RTC).<\/p>\n
In-house, but as-yet unverified, benchmarks show big efficiency gains for the E1 against Arm Cortex-M85 and Cortex-M33 parts. (\ud83d\udcf7: Efficient)<\/p>\n
It’s the efficiency that is the chip’s real selling point, though \u2014 naturally enough, given the company’s name. Efficient claims that between the Fabric architecture and a co-designed compiler dubbed effcc the Electron E1 can deliver an order order of magnitude energy efficiency gain \u2014 or more \u2014 over equivalent Arm parts. For matrix multiplication, Efficient claims a 94\u00d7 improvement over the Arm Cortex-M85 and 15\u00d7 over the more efficient Cortex-M33; for fast Fourier transformers, a 24\u00d7 and 13\u00d7 gain respectively; for convolution operations in computer vision, a whopping 322\u00d7 improvement over the Cortex-M85 and a still-impressive 29\u00d7 over the Cortex-M33.<\/p>\n