876 Neuromorphic Hardware for SNN LSTM
876 : Neuromorphic Hardware for SNN LSTM
- Author: Hunter Schweiger
- Description: Efficient neuromorphic hardware implementation of a Leaky Spike Neural Network (LSNN) with adaptive threshold and refractory period
- GitHub repository
- Open in 3D viewer
- Clock: 50000000 Hz
Neuromorphic Hardware for SNN LSTM
How it works
This LSNN (Leaky Spike Neural Network) implementation features:
- 12-bit membrane potential with configurable decay (DECAY_FACTOR = 1/4)
- Adaptive threshold mechanism with learning rate control
- 3-cycle refractory period after spike generation
- 7-bit spike counter for monitoring activity
- Base threshold of 100 units with dynamic adaptation
The design operates through several key mechanisms:
-
Membrane Dynamics:
- Integrates 8-bit input current
- Applies leaky decay of 1/4 per cycle
- Resets to 0 after spike
-
Adaptation Mechanism:
- Learning-enabled threshold adjustment (controlled by uio_in[0])
- Adaptation increases with each spike
- Gradual decay when not spiking
-
Output Monitoring:
- uo_out[7]: Refractory state indicator
- uo_out[6:0]: Current membrane potential
- uio_out[7]: Spike output
- uio_out[6:0]: Spike counter
How to test
Testing procedure:
-
Reset (rst_n = 0):
- Verify all state variables reset to 0
- Threshold should reset to base value (100)
-
Basic Operation:
- Apply input current through ui_in[7:0]
- Monitor membrane potential on uo_out[6:0]
- Observe spike generation on uio_out[7]
- Check refractory period indicator on uo_out[7]
-
Learning Mode:
- Set uio_in[0] to enable learning
- Verify threshold adaptation after spikes
- Monitor spike frequency changes
-
Performance Verification:
- Track spike count through uio_out[6:0]
- Verify proper threshold adjustment
- Test different input current levels
External Hardware
None required - all testing can be done through digital I/O
IO
| # | Input | Output | Bidirectional |
|---|---|---|---|
| 0 | Input current bit [0] | State variable bit [0] | |
| 1 | Input current bit [1] | State variable bit [1] | |
| 2 | Input current bit [2] | State variable bit [2] | |
| 3 | Input current bit [3] | State variable bit [3] | |
| 4 | Input current bit [4] | State variable bit [4] | |
| 5 | Input current bit [5] | State variable bit [5] | |
| 6 | Input current bit [6] | State variable bit [6] | |
| 7 | Input current bit [7] | State variable bit [7] | Spike bit [7] |