MG24 Enhances Intelligence and Energy Efficiency with MLPerf Tiny Benchmark
The combination of the Internet of Things (IoT) and Machine Learning (ML) is becoming increasingly popular in many applications, ranging from small IoT devices to large data centers. For various reasons, such as real-time/offline operations, battery savings, and even security and privacy considerations, it is ideal to run ML algorithms and data processing at the edge where the data is generated. These low-power IoT devices can be embedded microcontrollers, wireless SoCs, or smart devices integrated into sensor equipment.To achieve this goal, semiconductor manufacturers need small, low-power processors or microcontrollers to run the necessary machine learning (ML) software effectively, sometimes also supporting hardware accelerators built into the chip. Evaluating the performance of a specific device is not straightforward, as its effectiveness in ML applications is more critical than raw numerical computations.MLCommons™ (an open engineering alliance) has developed three benchmark suites to compare ML products from different vendors. MLCommons focuses on collaborative engineering efforts to benefit the machine learning industry through benchmarking, metrics, public datasets, and best practices.
These benchmark suites are called MLPerf™, which measure the performance of ML systems during inference when applying trained ML models to new data. The benchmarks optionally measure the energy used to complete inference tasks. Because the benchmarks are open-source and peer-reviewed, they provide an objective and fair test of performance and energy efficiency.Among the three benchmarks provided by MLCommons, Silicon Labs (also known as “Silicon Labs”) has submitted a solution for the MLPerf Tiny suite. MLPerf Tiny is aimed at the smallest low-power devices, typically used in deeply embedded applications such as IoT or smart sensing.Click the button at the end of the article to read the original text or visit the link below to explore related testing guidelines and results:https://cn.silabs.com/blog/machine-learning-benchmarks-compare-energy-consumptionSilicon Labs System BenchmarkingSilicon Labs submitted the EFR32MG24 multiprotocol wireless SoC for benchmarking. This SoC includes an Arm Cortex®-M33 core (78 MHz, 1.5 MB Flash / 256 kB RAM) and a Silicon Labs hardware accelerator subsystem. It supports multiple 2.4GHz RF protocols, including Bluetooth Low Energy, Bluetooth mesh, Matter, OpenThread, and Zigbee. It is an ideal choice for mesh IoT wireless applications such as smart homes, lighting, and building automation. This compact development platform provides a simple and time-saving path for smart and machine learning development.This SoC runs TensorFlowLite for Microcontrollers, which enables ML inference models to run on microcontrollers and other low-power devices with small memory. It utilizes optimized neural network kernels from the Silicon Labs Gecko Software Development Kit (SDK) in the CMSIS-NN library.Test ResultsMLPerf™Tiny v1.0 benchmark results highlight the efficiency of the Silicon Labs EFR32MG24 platform’s on-chip accelerator. Inference computations are offloaded from the main CPU, allowing it to perform other tasks or even enter sleep mode to save additional power.This is crucial for meeting the growing demand for ML-enhanced low-power wireless IoT solutions, allowing devices to operate for up to ten years on a coin cell battery. These latest results indicate a speed increase of 1.5 to 2 times compared to other benchmarks in the previous Tiny v0.7 results, with energy consumption reduced by 40-80%.As machine learning becomes more widely adopted in embedded IoT applications, this ability to run inference with low power is essential, enabling product designers to apply machine learning in new use cases and across different devices.Explore Silicon Labs‘s MG24 wireless SoC product information:https://cn.silabs.com/wireless/zigbee/efr32mg24-series-2-socsYou can also scan the QR code below to follow Silicon Labs on social media platforms.
