BrainChip Holdings Ltd., a leader in commercial production high-performance, ultra-low-power, event-based neuromorphic artificial intelligence platforms, said today it raised $25 million in new funding.

The company uses a brain-inspired hardware architecture to minimize the computations and data movement required by edge AI models. At the core of BrainChip’s offerings is Akida, its next-generation neuromorphic architecture for low-power devices. It supports numerous AI model subtypes, including convolutional neural networks, vision transformers, state-space models and more.

“Everybody wants their own variation, their own flavor, particularly companies building for specific use cases,” BrainChip Chief Executive Sean Hehir told SiliconANGLE in an interview. “Most people really don’t care — as long as you’ve got this breakthrough performance and it’s easy to implement.”

BrainChip offers licensed intellectual property for companies that want custom accelerators, while also delivering its own production-ready chips for customers without the resources to manufacture silicon themselves. The company further supports deployment through a catalog of machine learning models and state-space-model-based work tailored for edge inference.

Using BrainChip’s technology, the compact, cost-effective AKD1500 AI accelerator chip allows powerful models capable of learning and continuous improvement across vision, audio and sensor processing to run directly at the edge. The combination of high capability and ultra-low-power operation — continuously running on milliwatts — makes it possible to deploy always-on intelligence in wearables, sensors and battery-powered devices.

Not every customer can or wants to build custom silicon, Hehir explained. Producing BrainChip’s own chips widens its addressable market and supports customers who want to prototype real products before committing to full ASIC development.

“The world is going more and more AI-oriented… and people are pushing more and more compute away from the data center,” said Hehir.

Local-first generative AI and other model capabilities pave the way for robust capabilities that respect privacy and security by keeping all data on-device. They also create opportunities for long-running AI models in automobiles, drones and wearable devices such as smartwatches and pendants.

The company also offers easy-to-use Akida-based development planning, benchmarking, testing and validation in the company’s BrainChip Developer Akida Cloud. Software engineers and AI developers can quickly spin up a cloud instance to test their applications within minutes.

“I think our competitive advantage is we’re easy to adopt,” said Hehir. “I think two is performance, and three, we provide more value. We’re much more of a solutions provider.”

The new capital will enable BrainChip to execute a broader hardware roadmap: multiple versions of its AKD1500 accelerator, modular boards for PC-style integration, and complete reference designs for fast-growing edge-AI markets including radar, wearables and private voice assistants.

BrainChip’s technology is being adapted by multiple companies for ultra-low-power edge use cases.  HaiLa Technologies Inc. is adopting it for ultra-low-power Bluetooth and Wi-Fi integration for wearable visual classification. Deep Perception Inc. is developing a full visual pipeline for drones and mobile devices.

“We’re enabling other people’s products — the products that they make their money on,” Hehir said. “It’s all about adding value and making sure that they can make their products that much better.”

Image: SiliconANGLE/Microsoft Designer