Menu
开源 LFSR 随机泄漏积分-发放神经元硬件实现方法论 — SkyWater 130nm CMOS 工艺上的随机脉冲神经网络神经元设计
The review-first prompt is selected by default. You can switch to a direct command or download a local copy.
Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
npx skills add https://github.com/hiyenwong/ai_collection --skill lfsr-stochastic-lif-neuron-skywater-130nmThe command stays on one line. Scroll horizontally to inspect it before copying.
Prefer a local copy? Download the files currently available to SkillsMP.
Showing SKILL.md
Based on SOC occupation classification
| name | lfsr-stochastic-lif-neuron-skywater-130nm |
| description | 开源 LFSR 随机泄漏积分-发放神经元硬件实现方法论 — SkyWater 130nm CMOS 工艺上的随机脉冲神经网络神经元设计 |
| version | 1 |
| author | Kumaresan, Sivasubramani |
| arxiv | 2606.23532v1 |
| published | 2026-06-22T00:00:00.000Z |
| categories | ["cs.ET","cs.AR","cs.NE"] |
| keywords | ["stochastic spiking neuron","LIF","LFSR","neuromorphic hardware","SkyWater 130nm","open-source hardware"] |
| activation | 随机LIF神经元, LFSR神经元硬件, SkyWater neuromorphic, stochastic spiking hardware |
开源随机泄漏积分-发放(LIF)神经元硬件实现，使用线性反馈移位寄存器(LFSR)生成随机脉冲发放概率。在SkyWater 130nm CMOS工艺上实现，完全开源。
创新点：
LFSR驱动概率表：
LFSR → 8-bit comparator → Activation Table → Bernoulli(p) firing
周期特性：
饱和16位泄漏积分器：
参数配置：
面积：~10,600 μm² (70%利用率) 时序：50 MHz 正裕量 平台：Tiny Tapeout tile
问题：比较器输出在短滞后存在串行相关性，滞后8附近有负峰值
原因：比较字节每周期移位1位
解决方案：每16周期子采样恢复白化性
# 串行相关性检测
def check_serial_correlation(output_stream, max_lag=16):
autocorr = np.correlate(output_stream, output_stream, mode='full')
# 滞后8附近负峰 → 需要子采样
if autocorr[8] < 0:
return True # 需要16周期子采样
单调控制：
Rate Coding Sweep实验：
Weight sweep: [0, 255] → firing rate monotonic increase
Threshold sweep: [0, 65535] → firing rate monotonic decrease
Refractory: [0, 7] → rate cap at 1/(r+1) spikes/cycle
模型检查：
# cocotb测试框架
@cocotb.test()
async def test_lfsr_period(dut):
# 验证LFSR周期
period = await measure_lfsr_period(dut)
assert period == 65535 or period == 63
@cocotb.test()
async def test_activation_uniformity(dut):
# 验证激活表均匀性
distribution = await sample_activation_table(dut, n=10000)
assert uniformity_test(distribution)
问题：LFSR输出在短滞后存在相关性，影响脉冲发放的白性
解决：16周期子采样或选择更长周期的多项式
问题：不应期过短导致发放率过高，功耗激增
推荐：不应期≥3周期平衡性能和功耗
问题：未进行位精确RTL验证可能导致硬件行为与模型不符
必要：cocotb测试覆盖所有参数组合
# 18个cocotb测试覆盖:
- LFSR周期验证
- 激活表均匀性
- 泄漏积分器饱和
- 不应期正确性
- 串行接口功能
- 并行输入切换
https://github.com/santhoshs93/tt_um_santhosh_stoch_neuron225ce6e| Parameter | Value | Configurable |
|---|---|---|
| LFSR bits | 16 | Polynomial |
| Activation table | 8 entries | Yes |
| Integrator bits | 16 saturating | Threshold |
| Refractory | 0-7 cycles | Yes |
| Interface | 16 registers | Serial |