NXP Ara240 DNPU
AI Application Bring-Up and Optimization on Ara240 DNPU
We bring up vision and generative AI applications on Ara240 DNPU: model porting and quantization, custom kernels and graph optimizations, and the software ISP and real-time pipeline around the model, tuned to your accuracy and performance targets.
Who this is for
Evaluating Ara240
You are considering Ara240 for a product. We bring your model up on Ara240 and establish how it performs against your accuracy and performance targets, so the hardware decision is based on your own workload.
Building on Ara240
Ara240 is selected for your product. We bring your application up on Ara240 and tune it to your accuracy and performance targets.
Services
Model porting
When this applies
Your model was developed for another platform and now has to run on Ara240.
What we do
We port the model to Ara240 and bring it up as part of your application.
Quantization
When this applies
The model has to run at lower numerical precision on Ara240, and accuracy has to be maintained.
What we do
We quantize the model and measure its accuracy against your target.
Custom kernels and graph optimizations
When this applies
The model uses operations the standard toolchain does not cover, or runs short of your performance target.
What we do
We develop custom kernels for those operations and optimize the model graph for performance on Ara240.
Software ISP and real-time pipeline
When this applies
A vision model needs processed camera input, or the application has to run in real time.
What we do
We build the software ISP for camera input and the real-time pipeline around the model.
How the work fits with NXP's software
Our work builds on NXP’s software for Ara240 (the Ara SDK and eIQ® tools). The diagram shows what you provide, what NXP provides and where our work sits.
- Yours
- NXP
- Our work
Evaluating Ara240
Ara240 not yet selected
Building on Ara240
Ara240 selected for your product
Your vision or generative AI model
Built on NXP's software for Ara240 (Ara SDK, eIQ®)
Model porting
Quantization
Graph optimization and custom kernels
prepared model
Ara240 DNPU
runs the model
evaluatingbuilding
Your hardware decision
based on results with your own model
Integrated into your application
real-time pipeline
Your application on Ara240
Tuned to your accuracy and performance targets
How an engagement starts
Technical discussion
You describe your application, models and targets. We discuss what is involved and how it could be approached.
Written recommendation
We follow up with a written recommendation on the approach.
Scoped work
If you decide to proceed, the work is scoped and agreed with you.