Edge-side AI audio processor now features RISC-V AndesCore D25F

2 mins read

Andes Technology and Spacetouch Technology, a company specialising in consumer electronics, have jointly announced the AndesCore D25F has been adopted by the SPV60 series as an edge-side audio processor.

The D25F processor, which is based on the AndeStar V5 architecture, is a 32-bit RISC-V CPU IP core equipped with P-extension (DSP/SIMD extension, draft version). Combined with the RISC-V DSP/SIMD ISA compiler, DSP library and emulator, alongside the highly optimised AndeSoft NN Library, customers will be able to accelerate AI application computing.

In addition, the D25F has several optional features, such as instruction and data cache, low latency local memory, and ECC for memory protection.

It can be configured as an AXI 64-bit or AHB 64/32-bit bus interface, and there is also a port that can access local memory directly from outside. It can also be configured with a Platform Level Interrupt Controller (PLIC) that supports more than 1,000 interrupt sources for rapid interrupt response, prioritisation, and pre-emption.

Furthermore, the Andes Custom Extension (ACE) provides additional flexibility through customised special-purpose instructions.

At the same time, the new generation of SPV60 series edge-side AI audio processor chips launched by Spacetouch have adopted a CPU + NPU + uDSP heterogeneous multi-core architecture, combining different types of processors together to efficiently handle various tasks, especially in the field of audio processing.

The design of this heterogeneous multi-core architecture presents a significant improvement for AI audio processor chips.

The SPV60 series of on-device AI audio processors integrate the Andes D25F core and Spacetouch’s newly developed uDSP and AI NPU (Neural Network Processor). The D25F CPU has a maximum frequency higher than 400MHz and can handle general-purpose computing tasks. The NPU’s computing power reaches 100GOPs and is optimised for neural network computing.

The uDSP core focuses on digital signal processing that makes special parallel acceleration according to known requirements and algorithm accumulation.

The combination of these features enables better performance and efficiency in audio processing. The chip also has a high-performance audio AD (Analog to Digital) converter with a dynamic range of more than 105dB and THD+N less than -95dB; and a DA (Digital to Analog) converter with a dynamic range of more than 105dB and THD+N less than -90dB. It also has a 0V direct-drive headphone amplifier module with semi-professional performance.

The chip comes with USB2.0, SD, SPI, UART, I2C, I2S and other peripheral interfaces, and supports AI noise reduction, AI echo cancellation, AI acoustic noise suppression, speech recognition and other algorithms.

Spacetouch's edge-side AI audio processor chips also delivers low power consumption. This is critical for area sensitive use cases for embedded systems and mobile devices, as they can deliver high performance while extending the battery life of the devices.

Spacetouch provides professional development support including tools and reference designs, so that the SPV60 series chips can be widely deployed in intelligent voice, smart headphones, professional audio and other related fields. The chips have already entered the mass production stage.

"With the longstanding cooperation between Spacetouch and Andes Technology, our SoCs are equipped with AndesCore processors that offer powerful performance with multiple optional features to us,” said Hu Yingzhe, CEO of Spacetouch. "The combination of the D25F RISC-V processor and Spacetouch’s heterogeneous multi-core AI audio processor IC brings more innovation and possibilities to the fields of audio processing, embedded systems and mobile devices.”