Was 2020 the Year of Edge AI Compute?
In 2020, many companies released flagship low-power devices with the goal of bringing AI to the edge. We’ll review some of the most recent ones and see how they compare.
Last week, Synaptics released news of their Katana Edge AI platform, a combination of a power-efficient SoC and a software optimizer for AI. This news, while exciting, is yet another addition to the list of companies who released low-power processing systems in 2020. The goal for each: bring AI to the edge.
We’ll be taking a look at some of this year's more notable releases to see how they stack up and where the industry is headed.
Synaptics’ Katana Edge AI
The most recent of the products we’ll be looking at is Synaptics’ Katana Edge AI platform. Synaptics claims that in designing this platform, it aimed to address a growing industry demand for battery-powered AI devices, like portable speakers and the like.
Katana is purpose-built for "people or object recognition and counting, visual, voice or sound detection, asset or inventory tracking and environmental sensing." Image (modified) used courtesy of Synaptics
Katana marries two technologies: a low-power SoC and energy-efficient AI optimization software. The SoC is a multi-core processor architecture, consisting of proprietary neural network accelerators, domain-specific processing cores, and other features like on-chip memory. The software, on the other hand, leverages Eta Compute’s TENSAI Flow to introduce an AI compiler with performance- and power-optimized libraries, according to Synaptics.
While the news is recent, no hard specifications have been released yet on the performance of this system.
Maxim Integrated’s NN Accelerator Chip
Earlier this year All About Circuits had the privilege of talking with representatives from Maxim Integrated about a new product, the MAX78000, a neural network accelerator chip meant to enable AI in battery-powered IoT devices.
Overview of important functional blocks. Image used courtesy of Maxim Integrated
The new MAX78000 chip contains two low-power cores, the Arm Cortex-M4 core or a RISC-V core. It also consists of an FPU-based microcontroller and a convolutional neural network accelerator. Some noteworthy specs include a 400 times latency improvement on MNIST and a 1,100 time cut in energy consumption when running MNIST.
Compared to a low-power Cortex M4F, the device has 600 times lower energy consumptions during keyword spotting and a 200 times improvement on keyword spotting. Enthused, the folks at Maxim expressed that this was one of the biggest AI-focused innovations the company had released in a long time.
Mythic's AI Analog Matrix Processor
The editorial team at All About Circuits also recently spoke with Mythic CEO Mike Henry to see his company's unique approach to low-powered AI.
Mythic shook up the industry earlier this year with the M1108, the industry’s first AI analog matrix processor (AMP). The system works by combining true analog compute-in-memory with the high memory density of flash, delivering 8-bits of on-chip memory.
Mythic’s AMP is a dataflow architecture consisting of an array of 112 "tiles." Each tile contains an analog compute engine, a digital SIMD vector engine, a 32-bit RISC-V nano-processor, an NoC router, and a local SRAM. No external DRAM is required.
NVIDIA Xavier vs. AGX Mythic M1108. Image used courtesy of Mythic
The specs reveal that this device goes toe to toe with GPUs. Running ResNet-50 at its highest framerate, the M1108 has shown a peak of 35 TOPS, 870 fps, and only 4 W of power consumption. Compared to NVIDIA's Xavier AGX, the M1108 demonstrated better performance, less power consumption, smaller area, and cheaper price.
The Year of Edge Compute
The growing demand for portable, AI-powered devices has put the industry into hyperdrive in 2020, designing and releasing the next best low-power processing unit. Who would’ve thought we’d have analog AI in 2020?
While this article only highlighted three out of dozens of other edge AI-focused innovations, designers all over are finding ways to balance processing power, memory capacity, and size in these new devices.
Read Up on More Edge AI Innovation in 2020
Dozens of companies, including startups, have zeroed in on AI processing at the edge. Here are just a few from 2020 you may have missed.
- Edge AI Chips for Voice-Activated Devices Save Power, Protect Privacy
- AI at the Edge: New Machine Learning Engine Deploys Directly on Sensors
- Startup Claims Its AI Chips Can Outperform Google and Intel at Edge Computing
- Living on the Edge: A Look at Leading Chipmakers and Their Push for Intelligence
- Arm Cortex-M MCUs Can Get an Instant Machine Learning and Inference Facelift with NanoEdge AI
Have you worked on any edge AI designs this year? What hurdles did you overcome? Share your thoughts in the comments below.