MIT puts ICs and a network inside a sewable fibre

When you put it into a shirt, you can’t feel it at all. You wouldn’t know it was there,” said to professor Yoel Fink.

To create the fibre, MIT put the ~0.5mm 4bump chip-scale ICs into a 11 x 12 x 200mm preform along with the ends of four longitudinal tungsten wires.

Then, in a process that hardly seems possible, heated the preform close to melting and drew it into the 10m long fibre in a way which towed the wires in through the preform and pressed them against the solder bumps to connect all the ICs in parallel as the plastic hardened. 90% of the chips connected correctly.

Design of the preform required reverse-engineering the drawing process to predict how the hard chips would flow as the plastic softened and stretched out. To get it to work required plastics of different stiffness in the preform, precision placement of the chips (The use of a 0.125mm mill cutter in its making hints at the precision required) and a fifth (non-electrical) wire to help guide them during extrusion.

Physically, the finished fibre can take bending around a 12mm radius in its stride and the wires only break when the radius is reduced close to 3mm.

Electrically, there are memory (Microchip 24CW1280X eeprom) and temperature sensing (Maxim MAX31875) ICs in the wire, each with four bumps: two for power and two for the I2C bus. All electrical connections are via one end of the wire where an external STMicroelectronics STM32F401 handles processing – there are plans to deign a microcontroller that can be connected inside the fibre like the memory.

Several fibres have been made with different mixes of ICs. A 1m length of one type has 767kbit of memory – in a demonstration a 480kbyte file was stored for 2 months without power.

In another demonstration, a fibre was attached to a shirt and used to record several hours of temperatures along its length while the wearer exercised in various ways.

Enough data was gathered to externally train an exercise-identifying convolutional neural network, and this was compressed and loaded back into the memory chips of the fibre, after which it was 96% successful in identifying repeats of the various exercise types, according to ‘Digital electronics in fibres enable fabric-based machine-learning inference‘, a paper which describes the work in Nature Communications. This paper is well worth viewing as it is clearly written and explains the preform and fibre in detail.

MIT worked with the Rhode Island School of Design and Harrisburg University of Science and Technology.

Leave a Reply

Your email address will not be published. Required fields are marked *