This model maps weights to ternary values {-1, 0, 1} (aka trits). One trit holds log(3)/log(2) ≈ 1.58 bits of information. To represent a single trit by itself would require 2 bits, but it is possible to pack 5 trits into 8 bits. This article explains it well: https://compilade.net/blog/ternary-packing
By using 4 ternary weights per 8 bits, the model is not quite as space-efficient as it could be in terms of information density. (4*1.58)/8 = 0.79 vs (5*1.58)/8 = 0.988 There is currently no hardware acceleration for doing operations on 5 trits packed into 8 bits, so the weights have to be packed and unpacked in software. Packing 5 weights into 8 bits requires slower, more complex packing/unpacking algorithms.
My neighborhood has a vehicle gate that is opened by an RF clicker, and a pedestrian gate opened by an RFID tag. I copied both of mine to my flipper. A couple months ago the coin cell battery in my clicker died, but I had my flipper! I also use the cloned RFID tag fairly regularly. There are two RFID tags in the house and more than two people using them, so I use my flipper for that too.
By using 4 ternary weights per 8 bits, the model is not quite as space-efficient as it could be in terms of information density. (4*1.58)/8 = 0.79 vs (5*1.58)/8 = 0.988 There is currently no hardware acceleration for doing operations on 5 trits packed into 8 bits, so the weights have to be packed and unpacked in software. Packing 5 weights into 8 bits requires slower, more complex packing/unpacking algorithms.