Qualcomm Technologies Inc. is acquiring Edge Impulse Inc., a startup that helps developers run artificial intelligence models on connected devices.

The companies announced the deal today without disclosing the financial terms.

San Jose, California-based Edge Impulse raised $54 million from investors prior to the acquisition. It provides a cloud platform that companies can use to train AI models optimized for connected devices. According to Edge Impulse, its software is used by more than 170,000 developers worldwide.

AI models optimized to run at the edge are often trained on sensor data. For example, a neural network designed to detect when industrial equipment overheats might be trained on production line temperature measurements. Edge Impulse provides features for turning such measurements into AI training datasets.

The company’s platform includes tools that AI teams can use to create data pipelines. Those are software workflows that automate some of the work involved in assembling a training dataset from sensor readings. Such workflows can, for example, filter duplicate and erroneous records.

Edge Impulse’s platform also provides tools for performing feature engineering. This is the task of condensing raw sensor data into a form that can be more easily processed by AI models. A developer might, for example, turn a collection of temperature readings into a single average value that is simpler to analyze. 

After a software team creates a training dataset, Edge Impulse helps identify an AI model architecture suitable for the project at the hand. Connected devices often have limited processing capacity. Edge Impulse displays the hardware requirements of different AI architectures, which make it easier to find ones that are efficient enough to run on a company’s connected devices.

After developers train a custom AI model using Edge Impulse, the platform packages the algorithm into a C++ library. C++ is a programming language that offers better hardware-efficiency than Python, the go-to syntax for AI development. Edge Platform says it can reduce neural networks’ memory usage by more than 60%.

The company offers its development platform alongside a device called BrickML. It’s a compact, rectangular computing module designed to run AI models. Companies can attach it to industrial equipment, collect sensor readings and analyze the data locally using the onboard neural networks. 

“Edge Impulse gives developers a tool that automates data collection, simplifies model training, provides advanced optimization tools, and offers one-click deployment to many types of hardware,” co-founder and Chief Executive Officer Zach Shelby wrote in a blog post today. 

Qualcomm’s acquisition of Edge Impulse comes a few weeks after the chipmaker debuted Dragonwing, a processor portfolio optimized for connected devices. Some of the chips in the product line include an integrated graphics processing unit for running AI models. Others feature reliability optimizations that allow them to operate at sub-freezing temperatures.

Currently, Edge Impulse supports two Dragonwing processors: the QCS6490 and QCS5430. Both chips can be used to power ruggedized mobile devices, while the latter module also lends itself to robotics projects. Following its acquisition by Qualcomm, Edge Impulse plans to add support for more of the processors in the Dragonwing portfolio.

“In addition to support for Qualcomm Technologies’ hardware, we’re continuing support for edge hardware from our wide partner base, including MCUs, CPUs, GPUs and NPUs,” Shelby detailed.

Photo: Qualcomm