Digital devices are becoming ever more important in our lives — most of us carry at least one such device with us everywhere we go. That usually means remembering to grab that big brick of a cell phone you own, dragging it around and keeping tabs on it throughout the day, and pulling it out of a pocket each time it is needed. This routine has become so ingrained in us that most of us do not stop to think about why it is that we carry a phone everywhere, and if there may be a better way to deliver mobile computing resources.
A team of engineers at MIT have recently published a paper in Nature Communications that moves digital computing power to a form factor that requires no special effort or lifestyle change to make use of — clothing.
While this may sound like science fiction, the advancements presented are primarily simply in the manufacturing of the fibers, and in utilizing existing components in a novel way. They have developed a method to embed tiny off-the-shelf components, such as the sub-millimeter 24CW1280X digital memory chips from Microchip Technology, and 25 μm to 50 μm tungsten wires in a polymeric fiber. All components are embedded in a single soft, flexible fiber which can be sewn into existing articles of clothing.
While all chips on the fiber share data and control lines, they can be individually addressed using the I2C communications protocol. This means that all devices in a fiber can be accessed from a single connection port at its terminus. The researchers have been able to store up to 95 kilobytes of data per meter of fiber. That may not sound particularly impressive in our terabyte world, but with a little ingenuity, much can be accomplished even with such meager resources.
In one trial of the architecture, the team engineered a fiber containing both digital memory and digital thermistor devices to measure body temperature. This fiber was sewn into a shirt such that the thermistors were located under the armpits. In this way, measurements of the wearer’s body temperature were both captured and stored at 0.5 second intervals throughout the course of the day. These measurements were further associated with distinct activities of the wearer — sitting, standing, walking, and running.
The most interesting use of the technology flowed somewhat naturally from this data collection. A neural network was trained offline to associate temperature profiles with human activity. This convolutional neural network was compressed, and the resulting weights and biases, equations for feature selection, ReLu functions, and matrix multiplication code for the four layer, 1650 connection, neural network were all stored entirely in the memory cells of a digital fiber. Evaluation of the in-fabric neural network achieved 96.4% accuracy in classifying human activities. The team envisions this type of device being used not only to detect human activity, but also being used to detect temperature abnormalities, such as might occur in illness.
The fibers are proving to be durable, with tests showing them able to survive being washed more than ten times. Couple that with the large set of use cases that are possible, because many types of digital chips can be incorporated into the fibers, and this method may just see the light of day outside of a research lab in the near future. The researchers are looking forward to a day when their technique is used in applications related to autonomous drug delivery, neural interfaces, personal thermal management, and computing in fabrics.