From Crystal Clear Audio Capture to Power Efficient Computing: Actions Technology ATW6095 Empowers Mass Production of High End AI Note Devices

Hardware competition in the AI Note space is shifting from “being able to record audio” toward “clearer audio capture, more power‑efficient computing, and longer runtime”.

Actions Technology has long‑standing expertise in the AI Note segment. Its ATS308X chip family has powered a wide range of mainstream consumer products for office and education scenarios globally, including HONOR CHOICE KUMI AI note, Mobvoi TicNote for WeCom, Memoket Gem, and Xueersi xNote smart recorder, earning broad market recognition.

Recently, multiple high‑end AI Note devices built around Actions Technology’s new edge AI SoC, the ATW6095 series, have entered mass production and hit the market. End‑product models include YoooClaw C·ONE Pro by Yidao Intelligence and N.B.BOSS P2 Plus by Xinren Intelligence. Leveraging its in‑house compute‑in‑memory NPU, high‑quality multi‑microphone audio capture, and system‑level low‑power design, the ATW6095 series enables brand customers to balance complex‑scene sound pickup, local AI processing, and extended recording within tight battery budgets, helping high‑end AI Note products deliver differentiated, benchmark‑grade user experiences.

Power‑Efficient Computing: Compute‑in‑Memory NPU Enables Efficient Edge AI Execution

For device manufacturers, high‑end AI Note products have long faced a trade‑off between edge AI performance and battery life. Deploying more sophisticated local AI models drastically shortens continuous operating time; to preserve long runtime, developers are forced to scale down edge AI processing capabilities. The Actions ATW6095 series resolves this dilemma.

Equipped with Actions Technology’s proprietary compute‑in‑memory NPU, the chip moves computing units inside memory arrays, eliminating massive energy overhead caused by repeated data movement between compute and memory units in conventional architectures.

Thanks to this design, the ATW6095‑series NPU delivers 100 GOPS of single‑core computing power, with an energy efficiency ratio up to 6.4 TOPS/W under INT8 precision. It delivers tens‑fold higher energy efficiency compared with mainstream DSP and CPU solutions built on traditional architectures. This means the ATW6095 series consumes less power running identical AI inference workloads; given the same power budget, it can run larger, more complex AI algorithm models.

For demanding AI workloads requiring sustained computing performance above 10 GOPS — tasks too heavy for conventional DSPs — Actions Technology has established deep collaboration with Aizip, a leading US‑based edge AI algorithm vendor. Powered by the high‑efficiency compute‑in‑memory NPU of the ATW6095 series, the platform natively supports Aizip’s industry‑leading lightweight far‑field speech enhancement and sound‑source separation models.

The solution supports far‑field audio capture up to 10 meters away. In environments with diverse background noise, ASR character‑/word‑error rates can drop by as much as 60%. In directional sound‑pickup mode, it achieves up to 40 dB of interfering‑speech suppression, substantially boosting speech‑recognition accuracy. Going forward, supported by omnidirectional multi‑mic arrays, the platform will separate mixed speech from multiple simultaneous speakers into 2‑3 independent audio streams in real time, further cutting error rates. It delivers high‑quality audio input for cloud‑based ASR and large‑language models, unlocking more challenging AI use‑cases.

The ATW6095 series also integrates a 420 MHz HiFi5 DSP. Compared with legacy HiFi4 implementations, audio‑processing performance is improved by 200 %, while neural‑network processing capacity rises by 400 %.

The HiFi5 DSP works in close synergy with the compute‑in‑memory NPU, moving beyond simplistic hardware stacking. Latency‑sensitive front‑end audio capture tasks are offloaded to the DSP, while the NPU focuses on AI inference. The HiFi5 DSP also takes over supplementary computation for workloads not yet natively supported by the NPU but already software‑optimized. This well‑partitioned hardware collaboration fully enables core algorithms including beamforming, bone‑mic fusion, AGC, speech enhancement, noise reduction and directional sound pickup, unlocking full computing potential under constrained power envelopes.

Crystal‑Clear Capture: High‑Grade Audio Signal Chain for Challenging Acoustic Environments

Powerful AI cores and algorithms rely on pristine raw audio input. Built as a highly integrated single‑chip solution, the ATW6095 series unifies audio sensing and edge computation to form a complete capability loop.

The chip embeds Actions‑developed high‑performance ADC/DAC modules. The ADC achieves 112 dB SNR with ‑100 dB THD+N; the DAC reaches 113 dB SNR. Its single‑chip audio performance matches that of premium standalone audio codecs from global tier‑one suppliers, delivering low‑noise, low‑distortion, high‑fidelity raw audio streams that lay a solid foundation for accurate AI recognition.

On microphone interfaces, the ATW6095 series supports 4 AMIC inputs plus 8 DMIC inputs, exceeding typical industry specifications and offering abundant I/O options for multi‑microphone‑array designs. Brand customers can flexibly configure microphone count and placement for target use‑cases. Multi‑mic‑array setups enable robust operation in noisy, long‑distance and multi‑speaker scenarios.

Extended Recording: System‑Level Low‑Power Design for Prolonged Continuous Recording

Battery life is critical to user experience for always‑on AI Note devices. Real‑world testing shows that with a 4+1‑microphone configuration running multiple edge AI algorithms, the ATW6095 series keeps average operating current below 10 mA during recording, delivering more than 20 % power savings versus competing solutions under equivalent supply voltage.

Low‑power operation does more than extend battery runtime; it unlocks additional hardware capabilities.

The ATW6095 series integrates a 2D graphics‑acceleration engine and supports mainstream GUI frameworks such as LVGL. It enables local on‑device rendering of recording status, battery level, and audio waveform visualizations. The combination of graphics‑processing capability and low‑power architecture opens new product possibilities that balance intuitive user interaction and long‑duration recording.

Beyond AI Note: Expanding the AI Hardware Footprint

High‑end AI Note devices demand a sophisticated mix of sound acquisition, real‑time signal processing, neural‑network inference and system‑level power management.

Positioned for premium AI hardware markets, the ATW6095 series consolidates proprietary compute‑in‑memory NPU, HiFi5 DSP, high‑grade audio signal chain, and system‑level low‑power design into one single chip. This comprehensive feature‑set extends well beyond AI Note hardware. It is highly suitable for AI glasses, AI desktop assistant robots, and a broader spectrum of portable AI smart devices.

As AI interaction evolves toward deeper human‑machine collaboration, Actions Technology will keep advancing core technologies for edge AI computing, multi‑modal perception, and wireless connectivity. Through high‑efficiency edge AI chip solutions and open industry partnerships, Actions Technology empowers AI to go wherever users go.