Page 32 - EE Times Europe Magazine | April2019

32 EE|Times EUROPE

ARTIFICIAL INTELLIGENCE
Top 10 Processors for AI Acceleration

at the Endpoint

By Sally Ward-Foxton

hile the acceleration of artificial-intelligence and device includes an 800-MHz HiFi4 audio digital signal processor (DSP)
machine-learning applications is still a relatively new for pre- and post-processing of voice data.
field, there is a variety of processors springing up to
Waccelerate almost any neural network workload. From the XMOS xcore.ai
processor giants down to some of the newest startups in the industry, The xcore.ai is designed to enable voice control in artificial intel-
all offer something different — whether that’s targeting different verti- ligence of things (AIoT) applications. A crossover processor (with
cal markets, application areas, power budgets, or price points. Here is a the performance of an application processor and the low-power,
snapshot of what’s on the market today. real-time operation of a microcontroller), this device is designed for
machine-learning inference on voice signals.
APPLICATION PROCESSORS It is based on XMOS’s proprietary Xcore architecture, itself built
Intel Movidius Myriad X on building blocks called logical cores that can be used for I/O, DSP,
Developed by Movidius, the Irish startup that was bought by Intel in control functions, or AI acceleration. There are 16 of these cores on
2016, the Myriad X is the company’s third-generation vision- each xcore.ai chip, and designers can choose how many to allocate
processing unit and the first to feature a dedicated neural network to each function. Mapping different functions to the logical cores in
compute engine, offering 1 tera-operations per second (TOPS) of dedi- firmware allows the creation of a “virtual SoC,” entirely written in
cated deep neural network (DNN) compute. The neural compute engine software. XMOS has added vector pipeline capability to the Xcore for
directly interfaces with a high-throughput intelligent memory fabric to machine-learning workloads.
avoid any memory bottleneck when transferring data. It supports FP16 The xcore.ai supports 32-bit, 16-bit, 8-bit, and 1-bit (binarized)
and INT8 calculations. The Myriad X also features a cluster of 16 pro- networks, delivering 3,200 MIPS, 51.2 GMACCs, and 1,600 MFLOPS. It
prietary SHAVE cores and upgraded and expanded vision accelerators. has 1 Mbyte of embedded SRAM plus a low-power
The Myriad X is available in Intel’s Neural Compute Stick 2, effec- DDR interface for expansion.
tively an evaluation platform in the form of a USB thumb drive. It
can be plugged into any workstation to allow AI and computer-vision
applications to be up and running on the dedicated Movidius hardware
very quickly.
XMOS’s xcore.ai is based
NXP Semiconductors i.MX 8M Plus on a proprietary architecture
The i.MX 8M Plus is a heterogeneous application processor featuring and is designed specifically for
dedicated neural network accelerator IP from VeriSilicon (Vivante AI workloads in voice-processing
VIP8000). It offers 2.3 TOPS of acceleration for inference in endpoint applications. (Image: XMOS)
devices in the consumer and industrial internet of things, enough for
multiple object identification, speech recognition of 40,000 words, or
even medical imaging (MobileNet v1 at 500 images per second). AUTOMOTIVE SOC
In addition to the neural network processor, the i.MX 8M Plus fea- Texas Instruments Inc. TDA4VM
tures a quad-core Arm Cortex-A53 subsystem running at 2 GHz, plus Part of the Jacinto 7 series for automotive advanced driver-assistance
a Cortex-M7 real-time subsystem. For vision applications, there are systems (ADAS), the TDA4VM is TI’s first system-on-chip (SoC) with
two image signal processors that support two high-definition cameras a dedicated deep-learning accelerator on-chip. This block is based on
for stereo vision or a single 12-megapixel (MP) camera. For voice, the the C7x DSP plus an in-house developed matrix multiply accelerator

NXP’s i.MX 8M Plus is the company’s first application processor TI’s TDA4VM is intended for complex advanced driver-assistance
with a dedicated neural network accelerator. It’s designed for IoT systems that allow vehicles to perceive their environments.
applications. (Image: NXP Semiconductors) (Image: Texas Instruments Inc.)

APRIL 2020 | www.eetimes.eu