Mismatch-tolerant CMOS oscillator and excitatory synapse for bioinspired image segmentation

In this work, we propose a mismatch-tolerant CMOS oscillator and excitatory synapse for bioinspired image segmentation circuits using a modified LEGION algorithm. As a result of the combination of excitatory synapses (to synchronize coupled oscillators) and global inhibition (to separate unconnected objects) the oscillator array can segment the image objects by phase-encoding. However, the synchronizing mechanism is sensitive to frequency dispersion, and hence a mismatch-tolerant oscillator architecture is a key point for correct operation. With a feedback scheme based on switched capacitors and a large time constant low-pass filter, frequency variations are significantly reduced. The excitatory synapse circuit enables oscillator synchronization and is implemented by a quasi floating-gate MOS transistor. Transistor-level simulation results for CMOS 0.35 μm technology are presented to validate the functionality of the mismatch-tolerant oscillator and synapse.