Chair of Digital Communication Systems, Ruhr-University Bochum, Universitätsstraße 150, 44801, Bochum, Germany.
Department of Electrical and Computer Engineering, University of Virginia, Main Office: Room C210 Thornton Hall, 351 McCormick Road, PO Box 400743, Charlottesville, 22904, USA.
Sci Rep. 2024 Jul 18;14(1):16643. doi: 10.1038/s41598-024-67400-3.
Circuit implementations of neuronal networks so far have been focusing on synaptic weight changes as network growth principles. Besides these weight changes, however, it is also useful to incorporate additional network growth principles such as guided axon growth and pruning. These allow for dynamical signal delays and a higher degree of self-organization, and can thus lead to novel circuit design principles. In this work we develop an ideal, bio-inspired electrical circuit mimicking growth and pruning controlled by guidance cues. The circuit is based on memristively coupled neuronal oscillators. As coupling element, we use memsensors consisting of a general sensor, two gradient sensors, and two memristors. The oscillators and memsensors are arranged in a grid structure, where oscillators and memsensors realize nodes and edges, respectively. This allows for arbitrary 2D growth scenarios with axon growth controlled by guidance cues. Simulation results show that the circuit successfully mimics a biological example in which two neurons initially grow towards two target neurons, where undesired connections are pruned later on.
到目前为止，神经元网络的电路实现一直专注于突触权重变化作为网络增长原则。然而，除了这些权重变化之外，还可以采用其他网络增长原则，例如引导轴突生长和修剪。这些原则允许动态信号延迟和更高程度的自组织，从而可以产生新的电路设计原则。在这项工作中，我们开发了一种理想的、受生物启发的电模拟电路，该电路可以模拟由导向线索控制的生长和修剪。该电路基于由忆阻器耦合的神经元振荡器。作为耦合元件，我们使用由通用传感器、两个梯度传感器和两个忆阻器组成的 memsensors。振荡器和 memsensors 排列在网格结构中，其中振荡器和 memsensors 分别实现节点和边缘。这允许具有由导向线索控制的轴突生长的任意 2D 生长场景。模拟结果表明，该电路成功模拟了一个生物学示例，其中两个神经元最初朝着两个靶神经元生长，然后再修剪不想要的连接。
