Probabilistic Computing
Current neural network technology has shown immense progress, which led to great advancedment in AI and related research. However, these computing architectures tend to show dramatic increase in energy and time consumption as the computation difficulty increases, leading to NP-hard problems or un-solvable problems. To tackle these challenges, we propse probabilistic computing, a new computation paradigm that uses probabilistics bit to compute and handle data, which is more powerful in difficult and complex problems such as NP-hard, weather forecast etc.
We propose the use of impact ionization as the source of randomness. By utilizing iMOS, we can use current CMOS fabrication technology as well as 3-terminal or 4-terminal devices that are more robust to device variation. Furthermore, we can utilize different semiconductor channel materials such as III-V, 2D, topological etc to develop p-bits suitable for mass-production.
-
TFT/iMOS based P-bit development
Thin-film transistor(TFT) is a type of field-effect transistor(FET) which is manufactured by thin film deposition techniques. When a voltage is applied to the gate electrode, the channel is generated in the semiconductor layer. This allows current flow from the source to drain, enabling the transistor switch ON. TFTs has many applications such as integrated circuits, analogue switches, MOS sensor, etc.
Impact ionization MOS (I-MOS) is next-generation semiconductor which uses impact ionization as a carrier injection mechanism. Impact ionization is a carrier multiplication process in which a hole or electron with sufficiently high kinetic energy collides with a lattice atom and impact ionizes it into free holes and electrons. The generated electrons and holes further impact ionize more electron–hole pairs under the acceleration of an extra electric field, which eventually induces avalanche breakdown and rapid current multiplication. It is possible to obtain an ultralow subthreshold slope and higher ON-state current in I-MOS devices given that impact ionization is an abrupt and rapid function of the electric field [1].
I-MOS devices have various applications, including probabilistic computing, energy-efficient nano-electronic devices and spiking neural networks, due to their probabilistic behavior and steep switching characteristics.
[1] Zhai, Yongbiao, et al. "Energy-efficient transistors: suppressing the subthreshold swing below the physical limit." Materials Horizons 8.6 (2021): 1601-1617.