Actions Technology Achieves Breakthrough with 2nd-Gen In-Memory Computing NPU

Actions Technology has announced substantial progress in the development of its second-generation in-memory computing technology, marking a significant leap forward for its core computing engine designed for on-device AI applications.

Comprehensive Performance Leap Enables Larger Models

The new NPU architecture represents a generational upgrade over its predecessor, delivering optimized performance across several critical metrics.

  • Substantially Increased Computing Power: Provides a robust hardware foundation for running more complex neural networks.
  • Support for Larger Model Scales: Breaks previous limitations on model parameters, enabling more powerful AI applications directly on end devices.
  • Enhanced Energy Efficiency & Utilization: Boosts performance while optimizing power consumption—a key factor for battery-powered mobile devices.
  • Improved Quantization Accuracy: Helps maintain model performance when converting high-precision models to lower-bit formats, minimizing precision loss.

Clear Roadmap to Mass Production and Ecosystem Development

Leveraging this new technology, Actions has outlined a clear commercialization timeline. The first SoC integrating the second-gen in-memory computing NPU is scheduled for tape-out within this year, with mass production and adoption across multiple brands and product categories targeted for 2025.

To accelerate deployment, the company is concurrently investing in its AI development tool platform. This platform is designed to simplify the process for clients, enabling more efficient adaptation of AI models to Actions' hardware and fostering a cohesive ecosystem from chip to application.

Expanding Beyond Audio into Broader On-Device AI Markets

With the maturation of its second-generation core technology, Actions is strategically expanding its focus beyond its traditional stronghold in audio. The company is actively exploring new application scenarios for its on-device AI chips.

The long-term vision is to build a diversified ecosystem encompassing "multiple brands, product categories, and algorithms." This suggests potential applications ranging from smart headphones and home automation to various IoT devices requiring local AI processing. The goal is to transition on-device AI from a technical concept into tangible consumer experiences across a wide array of products.