Wearables are fast becoming a part of our daily lives. By the end of 2019, some 61 million people are expected to own at least one of these tiny devices that allow them to keep track of their physical activity and health, run applications and receive notifications remotely from a smartphone.
However being small has its drawbacks. Limited by their battery size, storage capacity and computational power, wearables usually need to be paired with a smartphone in order function effectively. But this may soon be a thing of the past with Stitch, a novel processor chip developed by NUS Computing researchers led by Professors Tulika Mitra and Peh Li-Shiuan.
“We are looking for solutions where the wearable device can do everything for you without having to rely on any other device. The challenge is achieving such high performance at low power on a very small device,” said Prof Mitra.
Most processor chips found in wearable devices today rely largely on software solutions to carry out their various functions. Stitch, on the other hand, makes use of both hardware and software solutions.
“Software is important as it gives a chip the flexibility to be used across a variety of different applications, whether it is to allow a user to interact with the wearable through finger gestures or to guide a user through his daily commute. Hardware is needed to drive the performance of a wearable device,” Prof Peh explained.
Stitch comprises 16 cores or processing units, put together on a mesh network. Instead of having a huge, complex hardware accelerator within each core which would bulk up the chip, each core has a tiny accelerator patch that are “stitched” together virtually to create large accelerators.
The ingenuity in the design also lies in the heterogeneity of the novel chip’s multiple cores. It contains patches with different functionalities, which allows each patch to function individually and follow its own set of instructions or computational patterns.
Despite being 7.5 times smaller, it performed two times better than a homogeneous-core chip developed by the team when they started the project in 2014. It chalked remarkably high scores when tested for various applications such as finger gesture recognition, image classification, and pinpointing user location.
“The benefit of such a hardware-software co-design is that you get low-power, high-performance across multiple application domains. We now can match the needs of each application to the patch with the corresponding functionality, hence enable the wearable device to perform effectively and independently,” Prof Mitra added.
However, the current version of Stitch is unable to perform multiple functions simultaneously, requiring the developer to pre-determine each task. Hence the next step, said Prof Mitra, is to make the chip more dynamic, optimising its performance.