Google introduced an open platform Coral NPU (Neural Processing Unit), which offers an open hardware accelerator for machine learning models and software tools for its use with standard AI engines. Coral can be used as the basis for creating energy-efficient systems on a chip (SoC) suitable for use in Internet of Things, edge computing and sensing boards, as well as ultra-low-power consumer wearables such as headphones, augmented reality glasses and smartwatches. The developments of the project are distributed under the Apache 2.0 license.
Coral NPU is aimed at running constantly running AI applications on the side of portable devices with minimal energy consumption. The baseline Coral NPU implementation delivers 512 billion operations per second (GOPS) performance while consuming only a few milliwatts of power. NPU designed for flexible modification architectures depending on the needs of SoC manufacturers. The first manufacturer to begin production of chips based on Coral NPU will be Synaptics, which announced the Astra SL2610 line of processors for Internet of Things devices, including the Torq NPU subsystem, implemented on the basis of the Coral NPU architecture.
Typical applications of Coral NPU include the use of AI for image and sound processing, user interaction and context awareness. For example, devices can run large language models and applications for facial and object recognition, visual search, speech recognition, live translation, speech transcription, highlighting keywords in speech, controlling gestures and voice commands, determining user activity (walking, running, sleeping) and type of environment (at home, on the street).
In NPU enabled 32-bit RISC-V RV32IMF_Zve32x instruction set architecture, AXI4 bus, four-stage instruction pipeline with in-order dispatch, out-of-order instruction completion, four-thread scalar and two-thread vector dispatch. The processor supports SIMD operations for simultaneous processing of 128-bit vectors and is equipped with 8 KB ITCM memory for instructions and 32 KB DTCM memory for data.
NPU includes three processors working together component: