Standards & Specifications

Updated Mar 2, 2026

Industry standards, specifications, and best practices for Edge AI development and deployment.

Model Exchange Formats

ONNX (Open Neural Network Exchange)

The most widely adopted open format for neural network model interoperability.

  • Maintainer: Linux Foundation (ONNX Community)
  • Specification: github.com/onnx/onnx
  • Use case: Framework-agnostic model sharing and deployment

TensorFlow Lite FlatBuffer Schema

Google’s format for mobile and embedded model deployment.

  • Maintainer: Google
  • Use case: TFLite ecosystem deployment

Performance Benchmarks

MLPerf Inference

Industry-standard benchmark for ML inference performance.

Safety & Reliability

ISO 21448 (SOTIF)

Safety of the Intended Functionality—relevant for AI in automotive.

  • Maintainer: ISO
  • Scope: Addresses risks from functional insufficiencies including ML
  • Standard: ISO 21448:2022

UL 4600

Standard for autonomous systems safety, including ML components.

  • Maintainer: UL (Underwriters Laboratories)
  • Scope: Safety case framework for autonomous products

EDGE AI FOUNDATION Standards

EAIF Model Card Specification

Foundation standard for documenting edge AI models.

  • Scope: Model metadata, performance characteristics, deployment requirements
  • Status: Draft v0.3

EAIF Hardware Benchmark Suite

Foundation benchmark for edge AI hardware evaluation.

  • Scope: Standardized workloads, measurement methodology
  • Working Group: Hardware Standards
  • Status: Under development

EAIF Privacy Assessment Framework

Guidelines for evaluating privacy properties of edge AI systems.

  • Scope: Data handling, on-device processing, federated approaches
  • Working Group: Privacy & Edge
  • Status: Under development