M.
2 AI Edge
accelerator module
The Axelera M.2 AI Edge accelerator module is a Uses Cases:
highly integrated accelerator solution targeting
computer vision inference applications.
Security
Handle multiple HD video
Key Features: streams, deliver complex
object detection and
tracking applications for
smart cameras and
embedded NVRs.
Powered by a single Metis AIPU – capable of
running multiple HD camera streams.
Industry 4.0
Deliver unprecedented
levels of performance for
Best-in-class energy efficiency of 15 TOPS/W machine vision,
– delivering server-class performance within robotics, and safety
applications in a
embedded power budgets. power-constraint
environment.
High performance quantization tools deliver
FP32 accuracy in an INT8 footprint – lowering Smart retail
system cost. Enable frictionless
shopping experience using
in-store computer vision
for loss prevention, shelf
analytics, AI-based POS,
and fully autonomous
Module is offered ready for integration; stores powered by just
delivers lowest TCO for edge deployments. walk out technology.
Others
Highest accuracy, lowest latency edge Healthcare, Agriculture,
inferencing solution. Drones/UAV, Logistic/AGV,
Transportation.
Technical
Specifications:
The Voyager™ SDK unlocks the full potential of
Axelera M.2 module creating a seamlessly
Form Factor: M.2 2280M
integrated platform.
State-of-the-art model zoo: image classification; object
Processor: 1x Axelera
detection, segmentation, keypoint detection, face
Metis AIPU
recognition
AI-Cores: 4 digital Library support for image processing: scaling; cropping;
in-memory compute cores normalization; format conversion; non-maximal
suppression (NMS)
On-chip SRAM: 16MB L1;
32MB L2 Import models from Pytorch, TensorFlow and ONNX
x86/ARM host support
Dedicated RAM: 512MB
Ubuntu/Yocto OS support
System interface: 4-lane TVM compiler includes: graph optimization and
PCIe Gen3 interface post-training quantization tools.
Thermal solution: Passive Customer models Model zoo Non-neural libraries
(when integrated)
Low-code deployment flow
Dimensions: Length: 80mm; Gstreamer TVM compiler Open CL/GL
(PyTorch, TF, ONNX) compiler
Width: 22mm; Height: 0.8mm
Pipeline Callback API
Operating System: Ubuntu
Customer application Inference server
20.04/22.04; Yocto
Metis runtime
Operating Temperature
PCle Ethernet
Range: -40 to +85°C
Metis hardware
Get in touch!
Ordering To order the M.2 AI Edge Accelerator Module, please visit:
information: [Link]/order-product
Part Number: AXE-METM2-00A000 Description: AIPU M.2 Edge accelerator module
