Great positive recent article on Android Guides (translated from Italian) outlining Pico benefits, workings and potential.
If the below is achievable with Pico you'd like to believe the company have a competitive product here that they absolutely should be able to get some sort of market penetration sooner than later.
Akida Pico: The NPU that powers artificial intelligence in battery-powered, low-power devices
- Akida Pico integrates advanced artificial intelligence into ultra-low power consumption devices, such as wearables and medical sensors.
- Its scalable and modular architecture allows you to adapt performance and energy consumption to the application.
- The MetaTF development environment facilitates the rapid implementation of AI models with full support for TensorFlow and PyTorch.
Isaac10/05/2025 05:19Updated on 10/05/2025 07:06
The progressive technological transformation driven by the’
artificial intelligence it is achieving unprecedented levels of efficiency and adaptability, especially in the low-power device industry. One of the great drivers of this revolution is BrainChip, a manufacturer that changed the rules of the game with
Akida Pico, One
ultra-low power NPU designed to bring AI to battery-powered devices and without constant access to traditional electricity networks.
Akida Pico: the revolution of ultra-efficient artificial intelligence for wearable devices
Akida Pico stands out for its
unique ability to perform AI tasks with power consumption less than one milliwatté. This level of efficiency leads to an unprecedented reduction in the typical power consumption of neural processing chips, allowing artificial intelligence to always be active even on devices with extremely low energy requirements.
Thanks to this innovation it is possible to integrate advanced artificial intelligence into
Smart headphones, smartwatches, health bracelets, autonomous sensors, small household appliances and medical devicesé. Artificial intelligence is therefore no longer exclusive to complex and connected systems, but is now present in a
discreet and continuous in the user's daily life through compact gadgets and wearable devices.
A big advantage of this NPU is that
It does not require external server support or constant Internet connectivity., freeing the device from dependence on other larger systems or the cloud. This guarantees real edge processing (edge AI), fundamental for applications where latency, privacy and power consumption are determining factors.
El
autonomous design The Akida Pico's capabilities allow it to function both as a standalone chip and as an integrated module into existing hardware platforms. His
minimum size of only 0.12 mm² it facilitates implementation even in smaller devices or whose architecture requires maximum compactness.
Versatility of design, manufacturing and integration
El
Akida Pico It is produced with technology
FDSOI at 22 nm (22FDX) by GlobalFoundries, one of the most efficient and convenient processes for chip manufacturing. Its architecture also allows it to be easily adapted or resized to even more advanced manufacturing processes, such as
12 nm and also 3 nm, depending on the needs of the supplements or the specific application for which the device is intended.
This degree of flexibility makes the NPU a perfect solution for both mass-produced products and high-end applications, allowing manufacturers to choose the ideal balance between cost and performance.
In its structure, the following are distinguished
150,000 transistors, which allow the execution of complex neural tasks such as
Voice activation, keyword recognition, noise suppression, presence detection, personal assistants and much moreé. All this without requiring large amounts of internal memory, as it can be completed with chips
SRAM external if the application requires it, going from the initial 0.12 mm² to just 0.18 mm² with 50 additional KB of memory.
For connection to the rest of the system components, Akida Pico incorporates a
DMA and an AXI bus matrix, allowing direct integration with microcontrollers, main processors and other functional units of the device. Its ability to directly access memory via interfaces such as AHB, eMMC and GPIO increases its adaptability in ultra-low power consumption and mixed architecture environments.
For all these reasons, Akida Pico is not only small and efficient: it is also
incredibly modular, which opens the door to its use in future intelligent devices which, for now, are still being conceptualized.
Main technical characteristics of the Akida Pico
- Consumption less than 1 milliwatt: guarantees continuous operation with batteries of at least 10 mAh, allowing the autonomy of sensors and equipment without the need for frequent recharging.
- Minimum total area: 0.12 mm² in basic configuration, expandable to 0.18 mm² by adding SRAM memory.
- Scalable production: Designed for 22nm lithographic processes, but adaptable to 12nm or more advanced technologies, even reaching 3nm for premium chips.
- Pure digital architecture: Allows the acceleration of specific models of neural networks, facilitating real-time and low-latency responses to key events and signal processing at the edge.
- Advanced interfaces: DMA connection, AXI bus, access to AHB and eMMC memories and easy integration with GPIO.
- Complete scalability: It can adapt to the energy and computing needs of any device, from simple home automation sensors to smart home appliances and advanced wearable medical devices.
This combination of technical benefits makes Akida Pico a true benchmark for the next generation of smart devices, promoting the democratization of integrated artificial intelligence where it was previously impractical due to energy or space limitations.
What differentiates Akida Pico from other NPUs?
In a context where artificial intelligence is often associated with large data centers or expensive hardware, Akida Pico represents a
radical changeé. Its low power consumption and high performance make it particularly attractive for manufacturers and developers looking for
Offer intelligent functionality in low-cost stand-alone devices or with extreme range requirements. which use low energy consumption components.
With Akida Pico it is possible for a
remote control, motion detector, environmental sensor or small IoT device they have advanced artificial intelligence capabilities, such as pattern recognition, noise filtering and intelligent responses to events, without increasing weight, complexity or production costs.
Thanks to
event-based architecture, the chip is able to filter false alarms, activating the host only when it detects a real event, significantly extending battery life and reducing the need to recharge or replace it.
This results in tangible benefits for the end user: more durable devices, less maintenance, greater privacy by no longer having to send data to servers, and a consistently smooth and reliable user experience.
Software development and compatibility platform
Development on Akida Pico is supported by
BrainChip's exclusive MetaTF software environment, which allows engineers to optimize and compile their own
Temporal neural networks (TENN) directly to the NPU ecosystem.
One of the great advantages is the compatibility of MetaTF with
TensorFlow, Keras and PyTorch, the most popular frameworks in the machine learning community. This means that developers can transfer their pre-trained models and deploy them to Akida Pico without having to learn new languages or methodologies, thus driving the rapid adoption and deployment of edge artificial intelligence projects.
The development environment offers:
- Easy model compilation and optimization for ultra-efficient hardware.
- Support for debugging, testing and training neural networks in real time.
- Documentation and examples to facilitate the inclusion of both artificial intelligence beginners and embedded systems experts.
- Ability to implement multimodal functions, which include vision, audio, presence detection and response to complex events.
Real and potential applications of the Akida Pico NPU
The capabilities of the Akida Pico allow it to be used in a number of solutions that were previously not equipped with artificial intelligence due to technical limitations. Among its most notable applications are
- Wearable devices and smart bracelets: Continuous health monitoring with real-time analysis and extended battery life.
- Device medicos: monitoring of vital signs, detection of abnormal events, warning systems and reduction of the number of recharges or battery replacements.
- Smart appliances: Refrigerators that recognize food, identify products close to the expiration date or detect the opening of doors using sensors and cameras integrated with local artificial intelligence.
- Environmental and home automation sensors: Home or building automation with pattern recognition, presence detection, event response and false alarm filtering.
- Personal headphones and speakers: Voice activation without Internet addiction, advanced noise suppression and personal assistants that work without latency or delay.
- Industrial and defense systems: Remote equipment monitoring, local data analysis, early fault detection and multimodal processing for critical applications.
- Robotics, drones and automotive: Quick responses, autonomous navigation, low latency and integrated processing without large on-board computers.
- New emerging devices: Technologies not yet developed will benefit from ultra-compact and efficient local AI, opening the door to innovation in the IoT and wearable markets.
Thanks to its insignificant consumption, it is possible to implement artificial intelligence functions in small
sensor hubs, alarm clocks, automatic doorbells, remote controls and other devices powered by batteries or even conventional batteries, where artificial intelligence was previously unthinkable.
Benefits for developers, integrators and users
- desarrolladores: Reduced learning curve thanks to the compatibility of MetaTF with standard ecosystems, simulators and rapid debugging tools.
- Supplements: Maximum flexibility to scale performance, combine it with other chips and choose the best balance between cost, performance and power consumption.
- Users: Smarter and more reliable devices, with greater autonomy and without the need to transfer sensitive data to the cloud, improving privacy and the overall experience.
The
scalability Another key point is the performance of artificial intelligence, as it allows you to modulate power and consumption based on the actual use and operating conditions of the device, maximizing battery life.
BrainChip's commitment to edge computing and integrated artificial intelligence is reflected in a roadmap that includes both proprietary hardware integration and software development, promoting a robust and collaborative ecosystem to expand artificial intelligence across devices of all types and functions.
The arrival of Akida Pico marks a turning point in the ability to transform any battery-powered device into an intelligent, autonomous device equipped with native artificial intelligence. All this without giving up autonomy, space or budget, and offering new opportunities both in the domestic sector and in professional, industrial and healthcare applications, laying the foundations for a new generation of connected and intelligent products.