NEXT-GENERATION PROGRAMMABLE SIGNAL PROCESSING IP

ACUity in Compute
Economy in Energy

ACU is a fully programmable real-time processing IP for audio-first, multimodal hearables and wearables. Positioned at the sensor I/O, it processes, synchronizes and fuses audio, biosignal, motion and lightweight visual streams before they reach system memory.

Explore the ACU Platform Explore Applications

1.3 µs

Input-to-output latency, down to

2 FP32 MACs

Per core, per cycle, sustained

mW-level

Continuous processing at the I/O boundary

Fully programmable

Streaming and block processing

THE ACU APPROACH

Fully programmable processing at the sensor I/O

ACU processes continuous sensor streams directly at the I/O boundary—before data enters system memory. Its fully programmable architecture handles audio, bio-signals and visual sensing as data arrives, reducing memory traffic, latency and energy.

Instead of forwarding every raw sensor stream to the central processor, ACU can output processed streams, extracted features, events and control metadata—reducing data volume before system memory.

Process. Synchronize. Fuse.

Sensor ADCs & Pre-processing SYSTEM MEMORY Pre-Processing in DSP SYSTEM MEMORY Large NN in NPU SYSTEM MEMORY Post-Processing in DSP SYSTEM MEMORY AUDIO EEG/EMG CAMERA IMU/OTHERS ACU: Fully Programmable Core Cluster Sensor Fusion & Pre/Post-Processing Core 1 Core 2 ··· Core N Sensor Fusion & Tiny NN (Sample) Tightly Coupled Shared Memory (Low Latency) SYSTEM MEMORY Large NN in NPU SYSTEM MEMORY

Traditional fragmented pipeline (top) vs. ACU sensor fusion cluster (bottom)

ACU can output Processed streams Extracted features Events Metadata & control

Process Before System Memory

Complete critical processing before data reaches system memory, reducing buffering, AXI traffic and unnecessary memory transfers.

Deterministic Sample-by-Sample Processing

Streaming execution delivers predictable input-to-output latency without relying on a software scheduler.

One Architecture, Multiple Signal Domains

Use the same programmable architecture for audio, bio-signals, visual sensing and tightly synchronized sensor fusion.

CORE CAPABILITIES

Built for continuous, real-time signal processing

Fully Programmable

Custom processing graphs, filters and algorithms—not fixed-function hardware.

Streaming + Block Processing

Sample-by-sample time-domain processing and block-based frequency-domain workloads.

Deterministic Ultra-Low Latency

Up to 768 kHz sample rate, with input-to-output latency down to 1.3 µs.

Synchronized Sensor Streams

Align audio, biosignal, motion and lightweight visual streams close to the sensor I/O, with support for PCM, PDM, DMA, memory streams and other sensor interfaces.

Runtime Reconfiguration

Seamless filter replacement and graph-level reconfiguration during operation.

Sustained FP32 Performance

Programmable RISC-V cores optimized for streaming signal processing, sustaining 2 FP32 MACs per core per cycle, including loads and stores.

PLATFORM

One Programmable Platform for Audio, Biosignals, Vision and Sensor Processing

The same ACU architecture supports real-time workloads across audio, biosignal, motion, lightweight vision and sensor-fusion domains.

Audio is the starting point. Multimodal sensing is the future. ACU connects them in real time.

REPRESENTATIVE WORKLOADS

Audio Processing

  • Programmable adaptive ANC and transparency
  • Hearing enhancement and beamforming
  • EQ, filtering and resampling
  • Voice and noise preprocessing
  • Audio feature extraction

Biosignal Preprocessing

  • EEG, EOG and EMG filtering
  • Artifact reduction
  • Frequency-band and envelope extraction
  • Event detection
  • Signal-quality monitoring

Vision & Other Sensors

  • Demosaicing and white balance
  • Line-buffer image preprocessing
  • IMU filtering and motion extraction
  • Sensor calibration and synchronization
  • Event detection and data reduction

Sensor Fusion

  • Audio + EEG / EOG
  • Audio + IMU
  • Audio + vision
  • Cross-sensor synchronization
  • Rule-based fusion and lightweight Tiny-NN inference

Built to Scale

The ACU architecture scales to match the workload, power, area, latency and integration requirements of each application and SoC.

Sensor inputs and interfaces
Processing throughput
Shared-memory capacity
Streaming workloads
Block-based workloads
Tiny neural networks
Power and area targets
Latency requirements
Optional accelerators

SOFTWARE & TOOLCHAIN

Programmability from architecture to deployment

The ACU SDK, graph tools and system-level simulator support development from algorithm mapping and static scheduling to performance estimation and deployment.

01

Describe the Processing Graph

Define audio, bio-signal and sensor-processing graphs through a portable interface.

02

Optimize & Validate

Optimize graph mapping, identify configuration issues and generate deterministic static schedules.

03

Simulate & Estimate

Evaluate functionality, latency and performance before silicon is available.

04

Generate & Deploy

Generate ready-to-load ACU code and integrate it into the target SoC.

ACU Simulator

System-Level Simulator

Model the complete signal-processing subsystem before tape-out to accelerate development, integration and customer support.

ACU Executable

Programmed in Standard C

All ACU kernels and libraries are written in standard C—no proprietary language or special tools required.

APPLICATIONS

One architecture for always-on sensing

From audio and bio-signals to visual and inertial data, ACU processes continuous sensor streams where latency, programmability and energy efficiency matter most.

Person wearing ANC headphones

Hearables & Hearing

Programmable, deterministic processing for adaptive ANC, transparency, hearing enhancement, hearing aids, cochlear implants, hearing protection, ear-EEG devices and sleepbuds.

Adaptive ANC Transparency Hearing Enhancement Hearing Aids Cochlear Implants Hearing Protection Ear-EEG Devices Sleepbuds
Sleep earbuds for overnight sensing

Biosensing & Health

EEG, EOG and EMG preprocessing with artifact reduction, feature and event extraction, and signal-quality monitoring—synchronized with audio and motion data.

EEG / EOG / EMG Artifact Reduction Feature & Event Extraction Signal-Quality Monitoring
Person wearing AI smart glasses

AI Glasses & Wearables

Always-on contextual sensing for AI glasses / eyewear and wearable assistants—lightweight visual and IMU processing for low-latency human-machine interaction.

AI Glasses / Eyewear AI Wearable Assistants Human-Machine Interfaces Always-On Contextual Sensing Lightweight Vision + IMU

SENSOR FUSION IN ACTION

Multimodal Signals Make Existing Algorithms More Context-Aware

Illustrative processing flows that customers and algorithm partners may implement on the ACU platform. They are not claims that ACUTE currently provides complete end applications.

Attention-Directed Interaction

EEG / EOG + Voice

Attention and gaze signals can help identify what the user is referring to, while voice provides the command.

Multimodal Hearing Enhancement

Audio + EEG / EOG + Head Motion

Cross-sensor context can help determine which speaker or sound the user intends to hear.

Context-Aware Adaptive Audio

Audio + Biosignals + Motion

ANC, transparency and hearing enhancement can adapt not only to the acoustic environment, but also to user context.

ABOUT ACUTE DESIGN

Deep Expertise Across the Complete Semiconductor IP Lifecycle

Founded in Grenoble, France, in 2025, ACUTE Design is a semiconductor IP company developing fully programmable processing architectures for continuous sensor data. Our team brings more than 80 years of combined experience across ultra-low-power processor architecture, RISC-V systems, IC design, embedded software, signal processing, and the development and commercialization of hearable, wearable and AIoT products.

From processor architecture and silicon implementation to software toolchains, system integration and customer deployment, the team covers the complete lifecycle of semiconductor IP development.

Processor & System Architecture

Ultra-low-power computing, RISC-V architectures, memory systems and multicore processing.

Silicon & IC Design

IC architecture, prototyping, verification, implementation and SoC integration.

Embedded Software & Toolchains

Embedded runtimes, graph-based development, simulation, optimization and deployment tools.

Audio & Wearable Systems

Programmable audio, hearables, wearables, AIoT applications and global customer deployment.