# Google Neural Network Models for Edge Devices: Analyzing and Mitigating Machine Learning Inference Bottlenecks

arXiv: https://arxiv.org/abs/2109.14320 · HTML: https://mlsanity.com/paper/2109.14320/

## Key facts

- Cited by 23 papers in the map (data through 14 Oct 2026).
- Builds on 45 of its 104 references that are in the map.
- First posted on arXiv 29 Sep 2021.

## Builds on (most cited first, 30 of 45)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [Very Deep Convolutional Networks for Large-Scale Image Recognition](https://mlsanity.com/paper/1409.1556/) (2014)
- [ImageNet Large Scale Visual Recognition Challenge](https://mlsanity.com/paper/1409.0575/) (2014)
- [Going Deeper with Convolutions](https://mlsanity.com/paper/1409.4842/) (2014)
- [Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation](https://mlsanity.com/paper/1406.1078/) (2014)
- [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://mlsanity.com/paper/1704.04861/) (2017)
- [Sequence to Sequence Learning with Neural Networks](https://mlsanity.com/paper/1409.3215/) (2014)
- [Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling](https://mlsanity.com/paper/1412.3555/) (2014)
- [Visualizing and Understanding Convolutional Networks](https://mlsanity.com/paper/1311.2901/) (2013)
- [Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation](https://mlsanity.com/paper/1609.08144/) (2016)
- [Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting](https://mlsanity.com/paper/1506.04214/) (2015)
- [Show, Attend and Tell: Neural Image Caption Generation with Visual Attention](https://mlsanity.com/paper/1502.03044/) (2015)
- [On the Properties of Neural Machine Translation: Encoder-Decoder Approaches](https://mlsanity.com/paper/1409.1259/) (2014)
- [ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices](https://mlsanity.com/paper/1707.01083/) (2017)
- [ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design](https://mlsanity.com/paper/1807.11164/) (2018)
- [SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size](https://mlsanity.com/paper/1602.07360/) (2016)
- [Deep Visual-Semantic Alignments for Generating Image Descriptions](https://mlsanity.com/paper/1412.2306/) (2014)
- [Show and Tell: A Neural Image Caption Generator](https://mlsanity.com/paper/1411.4555/) (2014)
- [In-Datacenter Performance Analysis of a Tensor Processing Unit](https://mlsanity.com/paper/1704.04760/) (2017)
- [Generating Sequences With Recurrent Neural Networks](https://mlsanity.com/paper/1308.0850/) (2013)
- [Long-term Recurrent Convolutional Networks for Visual Recognition and Description](https://mlsanity.com/paper/1411.4389/) (2014)
- [Unsupervised Learning of Video Representations using LSTMs](https://mlsanity.com/paper/1502.04681/) (2015)
- [Recent Advances in Convolutional Neural Networks](https://mlsanity.com/paper/1512.07108/) (2015)
- [LSTM: A Search Space Odyssey](https://mlsanity.com/paper/1503.04069/) (2015)
- [Streaming End-to-end Speech Recognition For Mobile Devices](https://mlsanity.com/paper/1811.06621/) (2018)
- [SCNN: An Accelerator for Compressed-sparse Convolutional Neural Networks](https://mlsanity.com/paper/1708.04485/) (2017)
- [Long Short-Term Memory Based Recurrent Neural Network Architectures for Large Vocabulary Speech Recognition](https://mlsanity.com/paper/1402.1128/) (2014)
- [Exploring Architectures, Data and Units For Streaming End-to-End Speech Recognition with RNN-Transducer](https://mlsanity.com/paper/1801.00841/) (2018)
- [Video (language) modeling: a baseline for generative models of natural videos](https://mlsanity.com/paper/1412.6604/) (2014)
- [Improving RNN Transducer Modeling for End-to-End Speech Recognition](https://mlsanity.com/paper/1909.12415/) (2019)

## Built on by (most cited first, 23 of 23)

- [High-performance deep spiking neural networks with 0.3 spikes per neuron](https://mlsanity.com/paper/2306.08744/) (2023)
- [An Exact Mapping From ReLU Networks to Spiking Neural Networks](https://mlsanity.com/paper/2212.12522/) (2022)
- [Special Session: Towards an Agile Design Methodology for Efficient, Reliable, and Secure ML Systems](https://mlsanity.com/paper/2204.09514/) (2022)
- [Accelerating Neural Network Inference with Processing-in-DRAM: From the Edge to the Cloud](https://mlsanity.com/paper/2209.08938/) (2022)
- [Shared Memory-contention-aware Concurrent DNN Execution for Diversely Heterogeneous System-on-Chips](https://mlsanity.com/paper/2308.05869/) (2023)
- [PIM-Opt: Demystifying Distributed Optimization Algorithms on a Real-World Processing-In-Memory System](https://mlsanity.com/paper/2404.07164/) (2024)
- [Machine Learning Training on a Real Processing-in-Memory System](https://mlsanity.com/paper/2206.06022/) (2022)
- [An Experimental Evaluation of Machine Learning Training on a Real Processing-in-Memory System](https://mlsanity.com/paper/2207.07886/) (2022)
- [SwiftRL: Towards Efficient Reinforcement Learning on Real Processing-In-Memory Systems](https://mlsanity.com/paper/2405.03967/) (2024)
- [EcoFlow: Efficient Convolutional Dataflows for Low-Power Neural Network Accelerators](https://mlsanity.com/paper/2202.02310/) (2022)
- [DeepSketch: A New Machine Learning-Based Reference Search Technique for Post-Deduplication Delta Compression](https://mlsanity.com/paper/2202.10584/) (2022)
- [Heterogeneous Data-Centric Architectures for Modern Data-Intensive Applications: Case Studies in Machine Learning and Databases](https://mlsanity.com/paper/2205.14664/) (2022)
- [Hardware-aware mobile building block evaluation for computer vision](https://mlsanity.com/paper/2208.12694/) (2022)
- [TransPimLib: A Library for Efficient Transcendental Functions on Processing-in-Memory Systems](https://mlsanity.com/paper/2304.01951/) (2023)
- [VEDLIoT -- Next generation accelerated AIoT systems and applications](https://mlsanity.com/paper/2305.05388/) (2023)
- [Harmonia: Enhancing Data Placement and Migration in Hybrid Storage Systems via Multi-Agent Reinforcement Learning](https://mlsanity.com/paper/2503.20507/) (2025)
- [A Study on Inference Latency for Vision Transformers on Mobile Devices](https://mlsanity.com/paper/2510.25166/) (2025)
- [NEON: Enabling Efficient Support for Nonlinear Operations in Resistive RAM-based Neural Network Accelerators](https://mlsanity.com/paper/2211.05730/) (2022)
- [Algorithm-hardware co-design for Energy-Efficient A/D conversion in ReRAM-based accelerators](https://mlsanity.com/paper/2402.06164/) (2024)
- [DCC: Data-Centric Compilation of Machine Learning Kernels for Processing-In-Memory Architectures](https://mlsanity.com/paper/2511.15503/) (2025)
- [T3C: Test-Time Tensor Compression with Consistency Guarantees](https://mlsanity.com/paper/2601.01299/) (2026)
- [Machine Learning-Driven Intelligent Memory System Design: From On-Chip Caches to Storage](https://mlsanity.com/paper/2603.14583/) (2026)
- [SISA: A Scale-In Systolic Array for GEMM Acceleration](https://mlsanity.com/paper/2603.29913/) (2026)

## Abstract (arXiv)

Emerging edge computing platforms often contain machine learning (ML) accelerators that can accelerate inference for a wide range of neural network (NN) models. These models are designed to fit within the limited area and energy constraints of the edge computing platforms, each targeting various applications (e.g., face detection, speech recognition, translation, image captioning, video analytics). To understand how edge ML accelerators perform, we characterize the performance of a commercial Google Edge TPU, using 24 Google edge NN models (which span a wide range of NN model types) and analyzing each NN layer within each model. We find that the Edge TPU suffers from three major shortcomings: (1) it operates significantly below peak computational throughput, (2) it operates significantly below its theoretical energy efficiency, and (3) its memory system is a large energy and performance bottleneck. Our characterization reveals that the one-size-fits-all, monolithic design of the Edge TPU ignores the high degree of heterogeneity both across different NN models and across different NN layers within the same NN model, leading to the shortcomings we observe. We propose a new acceleration framework called Mensa. Mensa incorporates multiple heterogeneous edge ML accelerators (including both on-chip and near-data accelerators), each of which caters to the characteristics of a particular subset of NN models and layers. During NN inference, for each NN layer, Mensa decides which accelerator to schedule the layer on, taking into account both the optimality of each accelerator for the layer and layer-to-layer communication costs. Averaged across all 24 Google edge NN models, Mensa improves energy efficiency and throughput by 3.0x and 3.1x over the Edge TPU, and by 2.4x and 4.3x over Eyeriss~v2, a state-of-the-art accelerator.