EventSpikeSeg: Memory-Aware Mixed-Signal Architecture for Event-Camera Spiking Segmentation
Organizations: Iran University of Science and Technology, Hengam Street, Resalat Square, Tehran 16846-13114, Iran
Abstract
Event-camera segmentation requires persistent neural state despite sparse input events, making memory organization a central constraint on embedded implementation. We present EventSpikeSeg, a mixed-signal architecture combining resident capacitor memory with shared analog processing in GF180MCU 180-nm CMOS. Its 25-population spiking encoder-decoder has 143,824 synaptic weights but requires 30,976,000 membrane states at 440 x 640 resolution. The mapping specifies chronological 5-ms model updates, digital event/address services and 64-state capacitor banks. Characterization covers 22 module families and 74 operating points, distinguishing transistor simulations from digital-library estimates. Nominal results include 41.711 microwatts biased neuron-core power, 3.338-pJ positive terminal energy for a 1.4-microsecond cell read, and 0.531-mV selected-state drift over a 5-ms extracted-tile hold. Commit and coupling errors are reported separately. The software checkpoint records 27.976% mean intersection-over-union on a held-out six-class DSEC split; this is not hardware accuracy. Component layouts and simulation evidence quantify local implementation costs and expose storage replication and service scheduling as scaling constraints. A detailed SPICE-calibrated behavioral forward simulation of one full-resolution DSEC frame achieves 24.26% mean intersection-over-union, versus 26.22% for the matched PyTorch reference, using a compact 5-fF bank variant and external count/voltage calibration. Physical full-network operation, system energy and mismatch robustness remain unverified.
Figures & tables
| Path / stage | Spatial size | Executed operations | Actors | Membranes |
| Native detail | Standard : , then residual . | 2 | 4,505,600 | |
| Half encoder | Standard , stride 2: ; residual standard and residual DW/PW at 32 channels. | 4 | 9,011,200 | |
| Quarter encoder | Standard , stride 2: ; DW/PW at 64 channels, then residual DW/PW. | 5 | 5,632,000 | |
| Eighth encoder | DW/PW stride 2: ; DW/PW at 128 channels, then residual DW/PW. | 6 | 3,097,600 | |
| Dilated context | Two residual DW/PW blocks at 128 channels; depthwise dilation 2. | 4 | 2,252,800 | |
| Decoder projections | Eighth / quarter | : at eighth; upsample and concatenate with quarter skip ( ); fuse , then project . | 3 | 1,971,200 |
| System dimension | Documented implementation and evidence boundary |
|---|---|
| Model execution | 25 PLIF actors, 54 directed edges; original float model and chronological DSEC validation. |
| Persistent storage | 30.976 million membranes, plus adaptation and class-current identities; native 64-cell storage banks in GDS. |
| Event communication | Digital packet/actor/destination controls, addressed analog contribution interfaces; bounded RTL/connectivity checks, not total workload throughput. |
| Neuron service | Shared restore, signed accumulation, leak, decision/reset, and commit modules; nominal module/tile transients, not all-state timing. |
| Dense readout | Six continuous class-current outputs; native readout module evidence, no full-hardware mIoU. |
| Physical result | Hierarchical GF180 research atlas; no compact die area, silicon measurement, or system-level PPA. |
| State category | Identities |
|---|---|
| PLIF membrane | 30,976,000 |
| Adaptation age | 10,982,400 |
| Adaptation amplitude | 10,982,400 |
| Continuous class current | 1,689,600 |
| Total | 54,630,400 |
| Element / operation | Window | Mean VDD power ( W) | VDD energy (pJ) | Characterization boundary |
|---|---|---|---|---|
| PLIF core, biased hold | 5 ms | 41.711 | 208,555 | One physical core, including standing bias. |
| PLIF update | 45 s | 41.603 | 1,872.15 | Injector parked; not full weighted update. |
| One resident cell, write | 1.2 s | Positive terminal imports: 0.04123 pJ. | ||
| One resident cell, read | 1.4 s | 2.382 | 3.3348 | Positive terminal imports: 3.33785 pJ. |
| Signed injector | 6 s | 0.14845 | 0.89068 | Pulse/recovery; independent phase generation excluded. |
| Resident adaptation | 200 s | 233.270 | 46,654.1 | Five age advances and two spikes. |
| Execution | mIoU (%) | Pixel accuracy (%) |
|---|---|---|
| PyTorch reference | 26.22 | 64.17 |
| SPICE-calibrated behavioral model | 24.26 | 61.30 |
| Difference (percentage points) |