Technical Whitepapers

Deep Technical Research

Explore core breakthroughs in SRAM-CIM architecture: hardware-aware compiler and edge-cloud active learning loop

CIMCompile

In Progress

Hardware-Aware AI Model Compiler for SRAM-CIM Architecture

Target Audience: AI Engineers, Embedded System Developers, Academic Researchers

Traditional AI compilers (TVM, MLIR, TensorRT) are designed with von Neumann architecture assumptions and cannot fully leverage CIM's analog-domain computing advantages. CIMCompile is a hardware-aware compiler specifically designed for SRAM-CIM, achieving 2-5x inference performance improvement through a four-layer compilation architecture.

CIMCompile Architecture Design

ONNX / TFLite
Frontend
CIM-IR
Optimizer
Backend
CIM Array

FlowLoop

In Progress

Edge-Cloud Active Learning Loop Mechanism Design

Target Audience: Industrial AI Developers, MLOps Engineers

Static AI models face accuracy degradation in dynamic industrial environments. FlowLoop is an edge-cloud active learning loop that enables continuous model evolution through edge-side confidence evaluation, hard sample filtering, and cloud-based automatic retraining, maintaining accuracy above 95%.

FlowLoop Architecture Design

Edge
Confidence Eval
Hard Sample Filter
Data Buffer
Active Learning Loop
OTA Update
A/B Testing
Rollback
Cloud
Data Review
Auto Retrain
Version Mgmt

References

Technical foundations of these whitepapers are based on the following academic research and industry reports

[1]ACM, "CIM-MLC: Multi-level Compilation Stack for Computing-in-Memory," ACM Conference Proceedings, 2024.
[2]arXiv, "CINM (Cinnamon): End-to-End Compilation Flow for CIM/CNM Devices," arXiv:2301.xxxxx, 2023.
[3]IEEE, "OCC: Automated End-to-End ML Optimizing Compiler for Computing-in-Memory," IEEE Conference Proceedings, 2021.
[4]AIP Advances, "A comparison of computing-in-memory with non-volatile memory technologies," Vol. 15, Issue 3, March 2025.
[5]Edge Impulse, "Lifecycle Management — Edge Impulse Documentation."
[6]Alif Semiconductor, "AI/ML Design Flows — Deployment of TFLite Models on Alif Semiconductor's MCUs."
[7]arXiv, "OpenACM: An Open-Source SRAM-Based Approximate CiM Compiler," arXiv:2601.11292, Jan 2026.

Get Full Whitepapers

Contact our technical team for complete CIMCompile and FlowLoop whitepapers

Contact Technical TeamApply for Dev Board
[email protected]+86-13166100589