Specifications
Product Family
AKD1500 M.2 module, AKD1500 PCIe development card and the BrainBoard1500 SPI module
Accelerator
BrainChip AKD1500 neuromorphic co-processor based on the Akida event-based architecture
Compute Fabric
Akida Neuron Fabric clocked from 5 MHz to 400 MHz
Performance
Up to 800 effective GOPS (800 GOPS at less than 300 mW)
Energy Efficiency
Better than 1 mW per GOP at maximum throughput
On-Chip Memory
1 MB local transfer memory
Learning
Adaptive on-chip learning — no cloud connection required
Host Interface
M.2 B+M key edge connector (module); PCIe development card for host evaluation
M.2 Form Factor
M.2 2230 (22 x 30 mm) with B+M key
Package
7 x 7 mm MFCTFBGA169, 0.5 mm ball pitch
Process
22 nm FD-SOI CMOS digital logic
Software
Industry-standard development environment with TensorFlow/Keras and PyTorch APIs
Model Library
BrainChip customer-ready models for keyword spotting, visual wake words and human activity recognition; TENNs and AkidaNet audio denoising and ASR models on request
Sensing Modes
Hear (audio and keyword spotting), See (vision object detection and classification), Sense (sensor anomaly detection)
Power Advantage
Enables fanless operation and a drop-in upgrade path without redesigning the power supply or cooling solution
Form Factor Benefit
The 2230 module frees M.2 slots typically reserved for memory and modems, so portable devices gain on-device AI without losing storage or wireless
Pricing
M.2 B+M key development card at 129 USD, BrainBoard 1500 SPI module at 99 USD, AKD1500 PCIe development card at 149 USD
Availability
M.2 module available now through the BrainChip web store; PCIe card launched September 2026
Target Markets
Aerospace, autonomous vehicles, robotics, industrial IoT, consumer devices and wearables
Overview
BrainChip's AKD1500 is a neuromorphic co-processor built on the company's event-based Akida architecture, and this card family is how it gets into real designs. The M.2 2230 module and the PCIe development card both deliver up to 800 effective GOPS from the Akida Neuron Fabric running between 5 and 400 MHz, into 1 MB of on-chip memory, at less than 300 mW — better than 1 mW per GOP at peak throughput. That power profile is what allows genuinely fanless operation and, critically, a drop-in upgrade path: engineering teams can add on-device AI to an existing product without touching the power supply or the thermal solution.
The M.2 2230 form factor is a deliberate design decision. At 22 x 30 mm with a B+M key edge connector it occupies the smallest M.2 socket class, freeing the larger slots normally used for memory and modems — which means tablets and other portable, connectivity-dependent devices can gain a neuromorphic accelerator without losing storage or wireless capability. The underlying silicon is a 7 x 7 mm MFCTFBGA169 package on a 22 nm FD-SOI CMOS process, and the architecture supports adaptive on-chip learning so devices can personalise without a cloud connection.
Software support comes through a mainstream development environment with TensorFlow/Keras and PyTorch APIs, plus customer-ready models for keyword spotting, visual wake words and human activity recognition, with TENNs and AkidaNet audio models available on request. BrainChip prices the family for evaluation rather than procurement friction: 129 USD for the M.2 development card, 149 USD for the PCIe card and 99 USD for the BrainBoard 1500 SPI module. QS Compute supplies BrainChip AKD1500 hardware and the wider neuromorphic and edge accelerator range; contact us for volume pricing and integration support.
Key Benefits
800 GOPS under 300 mW. Neuromorphic efficiency that makes fanless edge AI practical. M.2 2230 B+M key. Fits the smallest socket class and frees the slots used for memory and wireless. On-chip learning, no cloud. Adaptive personalisation at the device. Evaluates cheaply. Cards from 99 USD with TensorFlow/Keras and PyTorch APIs so teams can measure real power and latency before committing.
Applications
Always-on keyword spotting and voice wake in battery-powered products, visual wake words and low-power object classification, human activity recognition in wearables, sensor anomaly detection in industrial IoT, robotics and autonomous platforms needing event-based processing, aerospace and defence edge inference under strict power budgets, and any design where a conventional accelerator would breach the thermal or power envelope.
Request a Quote — BRAINCHIP AKD1500 — M.2 2230 AND PCIE EDGE AI DEVELOPMENT CARDS (800 GOPS UNDER 300 MW)
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →