- Admin
- November 21, 2024
Case Study
Edge AI - GUI based No-Code AI Platform
Client
A Tier-1 global semiconductor company that manufactures analog and embedded processing chips approached Ignitarium for an end-to-end AI / Deep Learning model development platform to help novice users train, compile and deploy such models on its embedded devices.
Challenges
While designing the solution, Ignitarium was required to address the following challenges:
• The platform was required to be deployed as a fully integrated no-code solution.
• In addition to deployment in the cloud, support for stand-alone deployment to customer labs.
• Simple interface and progress tracking to be provided for novice users to train and compile models for deployment into edge AI applications.
• Multi-tenant capability had to be built in for the platform with a constraint on resources availability for each user.
• Using the training and compilation software with pretrained models to train user’s own dataset by transfer learning and compile them for the customer’s various supported devices
• Providing accessibility to the edge device from a web browser with very low latency (2-3 seconds for live video streams).
• Integration and customization of the annotation tool to support the different types of labeling tasks
Solution
-
Ignitarium, through its deep expertise in Automotive processorbased designs, initiated a study of the RTL, DV coverage summary, and of the existing structural and functional FMEA reports to identify the gaps. The team also examined the HARA and specified safety goals.
-
DFMEA for IPs based on the component & safety requirements was prepared
-
Requirement traceability was updated. The Functional Safety Concept, Functional Safety Requirements and Technical Safety Requirements were defined
-
Next, the verification plan with requirement and specification traceability along with safety analysis and verification report were updated
-
Fault tree analysis (FTA) and its verification report were prepared
-
The RTL was then updated in compliance with functional safety, component verification infrastructure to achieve 100% coverage
-
A final design review was done by an external inspector to qualify the IPs to ISO 26262 ASIL-D
Fixed price engagement executed in three tranches
External inspector qualified the IPs to ISO 26262 ASIL-D
Business Impact
in just 7 months, as per ISO 26262 functional safety requirement
Ownership mode of execution with minimal intervention required from customer
Niche expertise in FuSA analysis, implementation and inspection all under one roof