"Low-Power Encrypted Computing Solutions"

The smart devices we use in our daily lives, such as smartphones, smartwatches, and smart health devices, generate large amounts of data. As the number of data sources such as these low-resource client devices continues to grow, the demand for sophisticated computing to extract value from the data using Machine Learning (ML) increases. Low-resource devices have limited computing capabilities due to their simple computing hardware and the energy limitation of their small batteries. These devices could use computational offloading in which sensor data is sent to a nearby edge device or the cloud for processing in order to get around these shortcomings. Offloading makes even very sophisticated data processing feasible, but only with the adjustment that the server performing the processing has unencrypted access to the data. Homomorphically encrypted computing is a new computing method aimed at mitigating these privacy concerns. It involves the client encrypting its data and sending the encrypted data for offloading. The offloaded processing occurs without decrypting the data. Although encrypted computing has a significantly high computational cost, advances in computer architecture and algorithms have made it possible to offload encrypted computation at a reasonable cost, thus making the technique feasible. However, these advances do not consider the costs posed to the low-resource client by encrypted computing, which make encrypted offload computing infeasible for low-resource devices. Therefore, researchers at Carnegie Mellon University (CMU) developed new algorithms and hardware designs to address these costs to client devices in order to make encrypted offloading possible for low-resource clients. This article continues to discuss the concept of homomorphically encrypted computing and the algorithms and hardware designs developed by the researchers to make encrypted offloading feasible for low-resource clients. 

CMU reports "Low-Power Encrypted Computing Solutions"