|Title||Edge Computing Hardware for Smart Buildings|
|Publication Type||Technical Report|
|Year of Publication||2019|
|Authors||Chai, E, Murmann, B|
|Keywords||Edge Computing Hardware, Hardware-Software Complexity Tradeoffs, Machine learning, Smart Buildings|
The state of the art in computer hardware has not kept up with the astonishing progress rate and requirements of deep learning and inference techniques for vision-based smart buildings. As a result, existing systems are not scalable, since they pipe unmanageable amounts of raw data from local sensors to remote servers. To prevent this data deluge, custom hardware for smart building solutions must be developed. The most suitable approach will process the raw data locally and forwards only relevant information to a central network that fuses this information for inference and action planning. We will leverage our expertise in hardware design to devise a custom solution for vision-based smart buildings. Our work will use prior work by Li & Milstein (Stanford CS & SoM) as a baseline. While we will initially work with semi-programmable chips (GPUs/FPGAs) to design and evaluate the local processing system, the long-term goal is to develop silicon chips that will be as efficient as the custom processors found in modern cell phones.