Abstract
The discovery of the memristor has opened new possibilities in the fields of electrical and computer engineering. Its inclusion as passive component in circuits enables for circuits to be designed that were not possible before. This paper shows the implementation of a decision tree as an analog circuit, including memristors to create an adjustable decision boundary. In conjunction with a circuit to control the memristor, a voltage scaling with the resistance is compared to a stream of input data. This circuit was designed with pixel-wise classification of images in mind. The implementation was tested with magnetic resonance images containing gliomas.It was evaluated whether the implementation of a decision tree in hardware is feasible, if the memristor is a suitable choice for this implementation and if this approach yields any benefit compared to equivalent software solutions. Performance wise the implemented prototype is in no way hindered by the memristor. The circuit is however overall not ideal as processing power and data access are too slow to compare to software implementations of the same algorithm on modern computers. The use of memristors as adjustable weight for classification algorithms is nevertheless a promising perspective.
| Original language | English |
|---|---|
| Title of host publication | 2019 IEEE 28th International Symposium on Industrial Electronics (ISIE) |
| Number of pages | 6 |
| Volume | 2019-June |
| Publisher | IEEE |
| Publication date | 06.2019 |
| Pages | 1357-1362 |
| Article number | 8781465 |
| ISBN (Print) | 978-1-7281-3667-7, 978-1-7281-3665-3 |
| ISBN (Electronic) | 978-1-7281-3666-0 |
| DOIs | |
| Publication status | Published - 06.2019 |
| Event | 28th IEEE International Symposium on Industrial Electronics - Pinnacle Hotel Harbourfront, Vancouver, Canada Duration: 12.06.2019 → 14.06.2019 Conference number: 150230 |
