IC-AKD1500S-FC020B-A
Showing
Overview
The AKD1500 SPI adds always-on AI with on-chip learning to MCU-based products over a standard SPI interface, so teams add AI without upgrading the MCU. The SPI peripheral runs in single, dual, quad or octal mode and needs no register setup after reset.
What's in the box
- AKD1500 Edge AI Co-Processor IC, SPI host interface
- Commercial temperature grade
- 1 or 5 chips, per selected package
Designed for
Battery-powered sensors and wearables
presence detection, vital-sign monitoring
Smart home and appliances
always-on keyword spotting and voice control
Industrial IoT nodes
vibration and anomaly detection at the sensor
MCU products
that need on-device personalization, with no cloud retraining
Overview
The AKD1500 PCIe connects to application processors over a PCIe Gen2 x2 endpoint, built for performance and scale. It delivers up to 800 effective GOPS at under 1 mW/GOPS, runs entire networks in its neural fabric, and can use host memory for larger models.
What's in the box
- AKD1500 Edge AI Co-Processor IC, PCIe host interface
- Commercial temperature grade
- 1 or 5 chips, per selected package
Designed for
Edge AI vision
ADAS/AV, vision-guided robotics, drones, video surveillance
Industrial PCs and edge servers
always-on anomaly detection and predictive maintenance
Multi-modal models
that need host memory and higher throughput
On-device personalized learning
with no cloud retraining
SPI or PCIe: at a glance
Select a column to switchField
Host Interface
SPI slave
PCIe Gen2 x2 endpoint
Clock Frequency
200 MHz
400 MHz
Power Dissipation
200 mW typical
300 mW typical
Form Factors
Chip only
Chip, M.2 cards, PCIe boards and systems
Field
Host InterfaceDiffers
SPI slave
PCIe Gen2 x2 endpoint
Clock FrequencyDiffers
200 MHz
400 MHz
Power DissipationDiffers
200 mW typical
300 mW typical
NPUs
32
32
On-Chip Memory
1MB SRAM
1MB SRAM
Temperature
Commercial, 0 to +70°C
Commercial, 0 to +70°C
Form FactorsDiffers
Chip only
Chip, M.2 cards, PCIe boards and systems
Architecture
Akida neuromorphic co-processor
Akida neuromorphic co-processor
Process
22nm FD-SOI CMOS digital logic
22nm FD-SOI CMOS digital logic
Package
7×7mm FCBGA169, 0.5mm pitch
7×7mm FCBGA169, 0.5mm pitch
Memory Expansion
SPI D/Q/O interface
SPI D/Q/O interface
On-Chip Learning
Yes (no cloud required)
Yes (no cloud required)
Software
BrainChip MetaTF: TensorFlow/Keras and PyTorch
BrainChip MetaTF: TensorFlow/Keras and PyTorch
Models
On GitHub: keyword spotting, edge learning, anomaly and fall detection
On GitHub: keyword spotting, edge learning, anomaly and fall detection
Compatible boards, cables, and reference platforms.
