研究背景
人体的复杂功能往往是通过使用简单的神经结构实现的。例如，自律神经系统通过交感神经和副交感神经的相互作用实现应急反应、压力效应和情绪调节，在无意识中协调哺乳动物器官的功能。作为神经系统中神经元之间的功能连接，突触可以通过释放不同的神经递质来传递信息。例如，去甲肾上腺素(NA)和乙酰胆碱(ACH)分别由交感神经和副交感神经的节后纤维释放。这两种递质在不同的节后神经元中产生不同类型的突触后电位，这是因为它们作用于不同的受体，通过相互转换、竞争和合作来调节心脏和瞳孔等器官。这种多重神经递质可以扩大突触权重的范围，包括在认知活动中有效地执行动态反应。
具有类似功能的人工突触需要在设备中设置不同的功能层来模拟不同神经递质的作用。一些高阶生理事件，特别是不受意识控制的行为，需要兴奋性突触和抑制性突触之间的协同作用。因此，需要一种能够模拟不同神经递质的释放机制和协调功能的人工突触。解决了这个问题，就可以开发一个单一的设备来模拟不同的环境反馈，并且可以拓宽突触可塑性和突触强度的形式，改善神经形态计算网络的功能，还可能简化电路的设计，从而可以用来模拟人类突触的丰富反应。对于可以模拟多种神经递质释放功能的人工突触，有以下项目需要改进：i) 需要进一步降低能耗，改善配对脉冲促进(PPF)，以实现高能效的神经形态计算，在突触上进行高权限的信息传递，并分别解码听觉或视觉信号中的时间信息，ii)长/短时可塑性(L/STP) 的界限模糊。而要实现绝对的STP，有必要在不植入离子的情况下引入功能层，iii)一些无意识控制的神经反射传递行为，如神经系统的功能模拟和一些器官的控制，需要进一步研究。
研究成果
自律神经系统通过复杂的神经通路维持生物体内的平衡，并对外部和内部环境的变化做出适应性反应。南开大学徐文涛教授团队报道了人工自律神经系统的制作，它复制了交感神经和副交感神经对心脏活动和瞳孔控制的综合影响，以模仿自律神经系统对外部变化的调节。人工自律神经控制瞳孔收缩和放松，调节正常心律和心律失常的心跳速度，这一点由信号指示器的眨眼率反映出来。这些功能是通过使用一个平行通道的突触晶体管进行切换的，该晶体管具有特殊的n-i-p异质结构，中间有一个二维h-BN绝缘体，以提供阻止离子注入二维MoS2底部n通道的屏障，实现乙酰胆碱诱导的短期可塑性，而顶部的P3HT纳米线p通道发生电化学掺杂反应，实现去甲肾上腺素诱导的相对长期可塑性。该装置显示了低至fJ的能量消耗和高达455%的超高配对脉冲促进指数。基于设备特性的人工神经网络实现了对心电图模式的高识别精度。这项工作扩展了对人工神经启发的生物信号处理和识别的见解。相关研究以“An Artificial Autonomic Nervous System That Implements Heart and Pupil as Controlled by Artificial Sympathetic and Parasympathetic Nerves”为题发表在Advanced Functional Materials期刊上。
研究亮点
1. 制作了一个人工自主神经系统，模拟交感神经和副交感神经对器官的联合作用，以控制人工瞳孔的收缩和放松，并直观地模拟正常和异常的心率。
2. 首次模拟了不同神经递质的共同释放，模拟了两种神经递质(NA 和ACH)的竞争和协同作用，并可以模拟静止和活动时的心率。
3. 这项工作为仿生学领域提供了潜在的应用，并为通过神经形态电子学模拟复杂功能提供了见解。
图文导读
Figure 1. a) Schematic of autonomic nervous system. Cholinergic neurons in sympathetic nerves of the thoracic spine release ACH to transmit signals, which can dilate the pupil and increase the heart rate. Adrenergic neurons in parasympathetic nerves in the brain release NA, which narrows the pupil and slows the heart rate. b) Schematic of the artificial autonomic nervous system, including parallel-channeled synaptic transistor, amplifying circuit, and Ni-Ti alloy artificial muscle fiber/signal indicators. The artificial muscle fiber is installed in artificial eyes to realize the dilation and contraction of pupils, and the signal indicator is installed on artificial heart models to illustrate the heart rate.
Figure 2. a) Schematic of biological synapse and the PCST with two independent channels, one for electrons and one for hole carriers, to simulate the transmission of two different neurotransmitters，e.g., NA and ACH. b) TEM micrograph of P3HT/PEO NWs printed on copper gauze. c) Elemental mapping of Sulfur in P3HT/PEO NWs. SEM image of d) P3HT/PEO NWs array and e) simple NW. f) AFM images of simple P3HT/PEO NW. STEM image of the lattice fringes of g) 2D MoS2 and h) h-BN. i) Room-temperature Raman spectra and photoluminescence spectra of 2D MoS2 nano-film.
Figure 3. a) Schematic diagram of transmission mechanism of neural synapse with two different neurotransmitters. b) Schematic diagram of PCST with independent channels for transporting holes and electrons in parallel. c) EPSC curves of pure MoS2 and h-BN MoS2 synaptic transistors triggered by 100 consecutive presynaptic spikes (4 V, 50 ms). d) EPSC for PCST triggered by different presynaptic spikes (4 or −4 V, 50 ms) under different VDS (1 and −1 V). e) Schematic diagrams of band bending in P3HT/PEO/ h-BN/ MoS2 junction under positive bias or negative bias of VDS and VGS. f) The spike-voltage dependent current characteristics triggered by presynaptic spikes from −4 to 4 V under VDS of −1 V and from 4 to −4 V under VDS of 1 V.
Figure 4. a) Schematic of rehearsal memory process of biological brain. b) EPSC curves for PCST triggered by 200 presynaptic spikes (−4 or 4 V, 50 ms), under VDS = 1 V. c) EPSC curves triggered by “dot” signals (4 V, 50 ms) and “dash” signals (4 V, 150 ms) for English letters of “NANKAI” in international Morse code. d) Emulation of classical conditioning learning on PCST. The presynaptic spikes of −3.5 and −4 V were used as mouse and hammer stimulus for conditioned learning， respectively. e) Schematic of NA and ACH release during exercising and resting state. f) EPSC curves and g) histograms mimicked regulations of nerve functions during and after exercise. Five consecutive negative (−2 V) and positive (3–4 V) presynaptic spikes were used to simulate the release of NA and ACH, respectively.
Figure 5. a) Analog weight update. Continuous increase of corresponding synaptic weight W caused by a series of consecutive negative spikes of −4 V, and decrease of W triggered by consecutive positive spikes of 2 or 4 V. b) Schematics of the five classes of ECG waveforms: N, S, V, F, and Q. c) Neuron network structure for pattern recognition. d) Recognition accuracy for ECG of Case #1 and Case #2 under different numbers of learning epochs. Inset: Confusion matrices of classification results between the output and the target ECG signals. e) Schematic diagram of parasympathetic and sympathetic nerve innervation to heart. f) EPSC curves and g) amplitudes mimicked regulations of normal cardiac rhythm and arrhythmia. Ten consecutive sets of negative (−2 V) and positive (3.5 V) presynaptic spikes were used to simulate the release of NA and ACH, respectively. The signal processing module biased and amplified the EPSCs, then converted them to voltage pulses to switch the signal indicators. h) Schematic illustration of artificial neuromuscular system: PCST, signal processing module and artificial muscle. i) EPSC of PCST triggered by negative and positive presynaptic spikes and the contractile force of a single artificial muscle; diameter of the artificial pupil at different time nodes (scale bar: 2 cm).
总结与展望
作者制作了一个人工自主神经系统，以模拟交感神经和副交感神经对器官的联合作用。这一目标是通过将一个平行通道的突触晶体管连接到人工肌肉上实现的。该晶体管使用了低维n-i-p无机/有机混合材料。人工自律神经被用来控制人工瞳孔的收缩和放松，并直观地模拟正常和异常的心率。对于人工突触，可利用平行通道突触晶体管同时传输电子和空穴载流子，并可瞬间切换，模拟不同神经递质(如NA和ACH)在突触裂隙中的传输。此外，首次模拟了两种神经递质的竞争和协同作用，协同作用调节了该心跳的速率，实现了静止和运动时的心率模拟。利用电流的增效和抑制的拟合结果模拟了心电图的识别。这项工作为设计和控制仿生学中的复杂功能提供了一种新方法。
文献链接
An Artificial Autonomic Nervous System That Implements Heart and Pupil as Controlled by Artificial Sympathetic and Parasympathetic Nerves
https://doi.org/10.1002/adfm.202210119
转自：“i学术i科研”微信公众号
如有侵权，请联系本站删除！
