Welcome Here  Shenzhen Mingjiada Electronics Co., Ltd.

With another wave of Deepseek, Actions Technology, in line with the trend of rapid expansion of artificial intelligence from cloud to end-side, as an important member of Actions Technology's end-side AI audio chip series, the ATS362X end-side AI chip is now officially released for AI entertainment audio devices, professional audio devices and AIoT edge computing terminals. ATS362X is now officially released. With its triple-core heterogeneous architecture, 24bit lossless sound quality and ultra-high energy efficiency ratio of 6.4 TOPS/W, the chip provides a powerful driving force for consumer and professional audio AI upgrades.

In November 2024, Actions Technology announced the launch of a new generation of end-side AI audio chip series based on Mixed-mode SRAM based CIM (‘MMSCIM’) technology.

ATS362X, as Actions Technology's new generation of tri-core heterogeneous SoC chips equipped with AI-NPU (Actions Intelligence NPU), adopts the ARM STAR CPU + HIFI5 DSP + MMSCIM NPU architecture design, which will inject AI power into a variety of end-side audio products, and help terminal brand products into the new era of AI. At present, the chip programme has entered the collaborative development stage with a number of brand customers, and is expected to achieve landing applications in terminal products soon.

Excellent AI computing power, empowering local real-time reasoning

As a masterpiece of Actions Technology's high-end AI audio chip, ATS362X adopts advanced highly flexible triple-core heterogeneous architecture, with HIFI5 and MMSCIM working in concert, and the theoretical arithmetic power reaches up to 132 GOPS, which supports real-time inference on the end side of the complex model such as voiceprint recognition, environmental sound classification, etc. The MMSCIM core has a high performance in the rarefied and rarefied environment, and is capable of real-time local inference. The MMSCIM core can be optimised from 6.4 to 19.2 TOPS/W@INT8 after sparse model optimisation, and the operating power consumption is only milliwatt level, which significantly improves the actual application arithmetic and energy efficiency ratio compared to traditional processor architectures. Meanwhile, the chip is fully compatible with mainstream AI frameworks such as TensorFlow, PyTorch, ONNX, etc., and provides ANDT toolchain and rich neural network operator libraries, which significantly improves the efficiency of model deployment. Meanwhile, Soft SIMD technology supports the development of customised operators, which can flexibly adapt to the cutting-edge needs of voice interaction, multi-modal perception, etc., and significantly expand the application boundaries of the product to provide a better environment for brand-name customers to carry out personalised AI application development. It provides comprehensive support for customers to develop personalised AI applications.

Professional-grade audio performance to meet high-fidelity demand

ATS362X has excellent audio performance. The chip integrates 4-channel 24-bit ADC (SNR>111dBA) and 2-channel 24-bit DAC (SNR>113dBA), with total harmonic distortion + noise (THD+N) as low as -100dB, making the sound quality comparable to that of professional recording equipment. The chip supports multi-channel I2S (16 channels @ 384KHz), ASRC dynamic sample rate adjustment (performance to THD+N less than -140dB), which enables accurate and error-free detail capture in complex sound fields, fully meeting the needs of high-fidelity scenarios such as professional stage speakers and digital mixing consoles.

Rich interfaces and security design to ensure product safety and reliability

In terms of connectivity, ATS362X integrates SPI (96MHz), USB 2.0 HS, SD/EMMC and other high-speed interfaces to meet the demand for multi-device interconnection and expansion. In terms of security design, the chip has built-in hardware-level AES-256/SHA-256 encryption, RSA2048 signature and TRNG random number generation, providing comprehensive protection for data privacy and device security.

The release of the ATS362X is an important milestone for Actions Technology in the area of end-side AI audio,’ said Gong Jian, general manager of Actions Technology's Portable Audio Business Unit. By integrating high-performance AI arithmetic, professional-grade audio processing capabilities and flexible interface design, we are committed to providing our customers with smarter, higher-quality and more personalised AI audio solutions, and to driving the entire industry to rapidly upgrade and iterate to enjoy the new era of AI.’