# Optimizing the Deployment of Tiny Transformers on Low-Power MCUs

arXiv: https://arxiv.org/abs/2404.02945 · HTML: https://mlsanity.com/paper/2404.02945/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- Builds on 35 of its 46 references that are in the map.
- First posted on arXiv 3 Apr 2024.

## Builds on (most cited first, 30 of 35)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://mlsanity.com/paper/1910.10683/) (2019)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [Hierarchical Text-Conditional Image Generation with CLIP Latents](https://mlsanity.com/paper/2204.06125/) (2022)
- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://mlsanity.com/paper/1704.04861/) (2017)
- [Gaussian Error Linear Units (GELUs)](https://mlsanity.com/paper/1606.08415/) (2016)
- [SSD: Single Shot MultiBox Detector](https://mlsanity.com/paper/1512.02325/) (2015)
- [FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning](https://mlsanity.com/paper/2307.08691/) (2023)
- [Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Coding](https://mlsanity.com/paper/1510.00149/) (2015)
- [Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention](https://mlsanity.com/paper/2006.16236/) (2020)
- [Rethinking Attention with Performers](https://mlsanity.com/paper/2009.14794/) (2020)
- [Linformer: Self-Attention with Linear Complexity](https://mlsanity.com/paper/2006.04768/) (2020)
- [GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints](https://mlsanity.com/paper/2305.13245/) (2023)
- [RWKV: Reinventing RNNs for the Transformer Era](https://mlsanity.com/paper/2305.13048/) (2023)
- [Fast Transformer Decoding: One Write-Head is All You Need](https://mlsanity.com/paper/1911.02150/) (2019)
- [Efficiently Scaling Transformer Inference](https://mlsanity.com/paper/2211.05102/) (2022)
- [A Survey of Transformers](https://mlsanity.com/paper/2106.04554/) (2021)
- [MCUNet: Tiny Deep Learning on IoT Devices](https://mlsanity.com/paper/2007.10319/) (2020)
- [Online normalizer calculation for softmax](https://mlsanity.com/paper/1805.02867/) (2018)
- [CMSIS-NN: Efficient Neural Network Kernels for Arm Cortex-M CPUs](https://mlsanity.com/paper/1801.06601/) (2018)
- [Data Movement Is All You Need: A Case Study on Optimizing Transformers](https://mlsanity.com/paper/2007.00072/) (2020)
- [The Devil in Linear Transformer](https://mlsanity.com/paper/2210.10340/) (2022)
- [Benchmarking TinyML Systems: Challenges and Direction](https://mlsanity.com/paper/2003.04821/) (2020)
- [Trained Quantization Thresholds for Accurate and Efficient Fixed-Point Inference of Deep Neural Networks](https://mlsanity.com/paper/1903.08066/) (2019)
- [DORY: Automatic End-to-End Deployment of Real-World DNNs on Low-Cost IoT MCUs](https://mlsanity.com/paper/2008.07127/) (2020)
- [Hydra Attention: Efficient Attention with Many Heads](https://mlsanity.com/paper/2209.07484/) (2022)

## Built on by (most cited first, 8 of 8)

- [From Tiny Machine Learning to Tiny Deep Learning: A Survey](https://mlsanity.com/paper/2506.18927/) (2025)
- [Automating Versatile Time-Series Analysis with Tiny Transformers on Embedded FPGAs](https://mlsanity.com/paper/2505.17662/) (2025)
- [Transformer-Based Fault-Tolerant Control for Fixed-Wing UAVs Using Knowledge Distillation and In-Context Adaptation](https://mlsanity.com/paper/2411.02975/) (2024)
- [Ariel-ML: Computing Parallelization with Embedded Rust for Neural Networks on Heterogeneous Multi-core Microcontrollers](https://mlsanity.com/paper/2512.09800/) (2025)
- [Future-Proofing Mobile Networks: A Digital Twin Approach to Multi-Signal Management](https://mlsanity.com/paper/2407.15520/) (2024)
- [Towards Real-Time ECG and EMG Modeling on μNPUs](https://mlsanity.com/paper/2604.18067/) (2026)
- [MR2-ByteTrack: CNN and Transformer-based Video Object Detection for AI-augmented Embedded Vision Sensor Nodes](https://mlsanity.com/paper/2605.15423/) (2026)
- [QTrans: A Quantum Transformer for Sentiment Classification](https://mlsanity.com/paper/2609.12011/) (2026)

## Abstract (arXiv)

Transformer networks are rapidly becoming SotA in many fields, such as NLP and CV. Similarly to CNN, there is a strong push for deploying Transformer models at the extreme edge, ultimately fitting the tiny power budget and memory footprint of MCUs. However, the early approaches in this direction are mostly ad-hoc, platform, and model-specific. This work aims to enable and optimize the flexible, multi-platform deployment of encoder Tiny Transformers on commercial MCUs. We propose a complete framework to perform end-to-end deployment of Transformer models onto single and multi-core MCUs. Our framework provides an optimized library of kernels to maximize data reuse and avoid unnecessary data marshaling operations into the crucial attention block. A novel MHSA inference schedule, named Fused-Weight Self-Attention, is introduced, fusing the linear projection weights offline to further reduce the number of operations and parameters. Furthermore, to mitigate the memory peak reached by the computation of the attention map, we present a Depth-First Tiling scheme for MHSA. We evaluate our framework on three different MCU classes exploiting ARM and RISC-V ISA, namely the STM32H7, the STM32L4, and GAP9 (RV32IMC-XpulpV2). We reach an average of 4.79x and 2.0x lower latency compared to SotA libraries CMSIS-NN (ARM) and PULP-NN (RISC-V), respectively. Moreover, we show that our MHSA depth-first tiling scheme reduces the memory peak by up to 6.19x, while the fused-weight attention can reduce the runtime by 1.53x, and number of parameters by 25%. We report significant improvements across several Tiny Transformers: for instance, when executing a transformer block for the task of radar-based hand-gesture recognition on GAP9, we achieve a latency of 0.14ms and energy consumption of 4.92 micro-joules, 2.32x lower than the SotA PULP-NN library on the same platform.