Mannhee Cho, Minil Kang, Minseong Um, Hangue Park, Hyung-Min Lee
{"title":"基于互反抑制的自振荡神经网络局部膜动态偏置CMOS LIF神经元。","authors":"Mannhee Cho, Minil Kang, Minseong Um, Hangue Park, Hyung-Min Lee","doi":"10.1109/TBCAS.2025.3583093","DOIUrl":null,"url":null,"abstract":"<p><p>This paper presents a CMOS-based neuron network that can emulate self-oscillatory biasing behaviors found in biological neural oscillator models. Based on leaky integrate-and-fire (LIF) neuron models, the proposed neuron circuit adopts the concept of reciprocal inhibitory network and synaptic fatigue as well as excitatory drive stimulation for replicating extracellular fluidic biasing of membrane potentials. On top of the base neuron circuit, an excitation integrator integrates positive and negative excitatory input spikes to stimulate the membrane potential bias, and a bias controller receives inhibitory drive input and generates output inhibitory drives depending on the membrane potential bias level. The proposed networks of multiple neurons with inhibitory connections can generate oscillating membrane potential biases, which can be used as local dynamic thresholds for neuron spike firing, resulting in self-patterned output spikes such as switching or dynamic firing rate patterns. The proposed neuron network was implemented with 250-nm CMOS process operating at the supply voltage of 2.5 V and consuming average power of 99.31μW per neuron during full operation. Operation waveforms were measured in various input conditions which can produce multiple output patterns. Variances in output signals due to process variation were measured from 32 neurons to verify the stability of operation, showing the standard deviation of 18% in the membrane potential gain per input spike and 12% in oscillation periods of the membrane potential bias. The results verified that the proposed neuron network can replicate the self-oscillatory behaviors of biological neuron models.</p>","PeriodicalId":94031,"journal":{"name":"IEEE transactions on biomedical circuits and systems","volume":"PP ","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2025-06-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"CMOS LIF Neurons with Local Membrane Dynamic Biasing Based on Reciprocal Inhibition for Self-Oscillatory Neural Networks.\",\"authors\":\"Mannhee Cho, Minil Kang, Minseong Um, Hangue Park, Hyung-Min Lee\",\"doi\":\"10.1109/TBCAS.2025.3583093\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p><p>This paper presents a CMOS-based neuron network that can emulate self-oscillatory biasing behaviors found in biological neural oscillator models. Based on leaky integrate-and-fire (LIF) neuron models, the proposed neuron circuit adopts the concept of reciprocal inhibitory network and synaptic fatigue as well as excitatory drive stimulation for replicating extracellular fluidic biasing of membrane potentials. On top of the base neuron circuit, an excitation integrator integrates positive and negative excitatory input spikes to stimulate the membrane potential bias, and a bias controller receives inhibitory drive input and generates output inhibitory drives depending on the membrane potential bias level. The proposed networks of multiple neurons with inhibitory connections can generate oscillating membrane potential biases, which can be used as local dynamic thresholds for neuron spike firing, resulting in self-patterned output spikes such as switching or dynamic firing rate patterns. The proposed neuron network was implemented with 250-nm CMOS process operating at the supply voltage of 2.5 V and consuming average power of 99.31μW per neuron during full operation. Operation waveforms were measured in various input conditions which can produce multiple output patterns. Variances in output signals due to process variation were measured from 32 neurons to verify the stability of operation, showing the standard deviation of 18% in the membrane potential gain per input spike and 12% in oscillation periods of the membrane potential bias. The results verified that the proposed neuron network can replicate the self-oscillatory behaviors of biological neuron models.</p>\",\"PeriodicalId\":94031,\"journal\":{\"name\":\"IEEE transactions on biomedical circuits and systems\",\"volume\":\"PP \",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2025-06-25\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"IEEE transactions on biomedical circuits and systems\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1109/TBCAS.2025.3583093\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"IEEE transactions on biomedical circuits and systems","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/TBCAS.2025.3583093","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
CMOS LIF Neurons with Local Membrane Dynamic Biasing Based on Reciprocal Inhibition for Self-Oscillatory Neural Networks.
This paper presents a CMOS-based neuron network that can emulate self-oscillatory biasing behaviors found in biological neural oscillator models. Based on leaky integrate-and-fire (LIF) neuron models, the proposed neuron circuit adopts the concept of reciprocal inhibitory network and synaptic fatigue as well as excitatory drive stimulation for replicating extracellular fluidic biasing of membrane potentials. On top of the base neuron circuit, an excitation integrator integrates positive and negative excitatory input spikes to stimulate the membrane potential bias, and a bias controller receives inhibitory drive input and generates output inhibitory drives depending on the membrane potential bias level. The proposed networks of multiple neurons with inhibitory connections can generate oscillating membrane potential biases, which can be used as local dynamic thresholds for neuron spike firing, resulting in self-patterned output spikes such as switching or dynamic firing rate patterns. The proposed neuron network was implemented with 250-nm CMOS process operating at the supply voltage of 2.5 V and consuming average power of 99.31μW per neuron during full operation. Operation waveforms were measured in various input conditions which can produce multiple output patterns. Variances in output signals due to process variation were measured from 32 neurons to verify the stability of operation, showing the standard deviation of 18% in the membrane potential gain per input spike and 12% in oscillation periods of the membrane potential bias. The results verified that the proposed neuron network can replicate the self-oscillatory behaviors of biological neuron models.