Tiny Tapeout > Tiny Tapeout Chips > Tiny Tapeout IHP 26b > 360 Neurona LIF con Aprendizaje STDP Dinamico (IEEE)
360 Neurona LIF con Aprendizaje STDP Dinamico (IEEE)
360 : Neurona LIF con Aprendizaje STDP Dinamico (IEEE)
- Author: Casandra Saray Galiano Quillahuaman
- Description: Neurona artificial biologica con plasticidad configurable y monitoreo de peso en tiempo real. — Equipo Triada CEG
- GitHub repository
- Open in 3D viewer
- Open in Tiny Tapeout Explorer
- Clock: 10000000 Hz
How it works
The design implements a biologically-inspired Leaky Integrate-and-Fire (LIF) neuron with a digital STDP (Spike-Timing-Dependent Plasticity) learning core:
stdp_core: tracks the time elapsed since the last pre-synaptic (spike_pre) and post-synaptic (spike_post) spikes using two down-counters (tau_pre,tau_post), each reloaded to a configurablewindow_sizeon a spike. If the post-synaptic neuron fires whiletau_preis still counting down, the synapticweightis potentiated (LTP). If a pre-synaptic spike arrives whiletau_postis still counting down, the weight is depressed (LTD). Weight is a saturating 6-bit register (0-63), initialized to 32.neuron_lif: integrates an 8-bit membrane potentialv_memby adding the (weight-scaled)stimuluseach cycle and subtracting a configurableleak. Whenv_memcrosses the configurablethreshold(scaled to 8 bits), the neuron emits onespikepulse and resetsv_memto 0.- Top level (
project.v): wires an external input pulse (ui_in[0],spike_pre) into both the STDP core and, when active, into the LIF neuron's stimulus (scaled by the current synaptic weight). The LIF neuron's own output spike closes the loop back into the STDP core asspike_post, so the weight adapts based on the relative timing between the external input and the neuron's own firing.
How to test
Configure the neuron via ui_in and uio_in, then drive spike_pre pulses on ui_in[0] and
observe spike_post and the live synaptic weight on uo_out:
ui_in[0]: external input spike (spike_pre)ui_in[4:1]: leak configurationui_in[7:5]: STDP time windowuio_in[3:0]: firing thresholduo_out[0]: output spike (spike_post)uo_out[6:1]: current synaptic weight (0-63)
test/test_project.py drives repeated input pulses at a short interval to force potentiation
(LTP), then lowers the threshold to force the neuron to fire ahead of the input and drive
depression (LTD), logging the synaptic weight after every pulse.
External hardware
None - this project only exercises the dedicated and bidirectional I/O pins directly.
IO
| # | Input | Output | Bidirectional |
|---|---|---|---|
| 0 | spike_pre | spike_post | threshold[0] |
| 1 | leak[0] | weight[0] | threshold[1] |
| 2 | leak[1] | weight[1] | threshold[2] |
| 3 | leak[2] | weight[2] | threshold[3] |
| 4 | leak[3] | weight[3] | |
| 5 | stdp_window[0] | weight[4] | |
| 6 | stdp_window[1] | weight[5] | |
| 7 | stdp_window[2] | unused |