Until now, neuromorphic calculations have tried to imitate the synapses between neurons in the brain. However, the new approach presented at the IEEE International Conference is aimed at acting like dendrites, subtle structures that move away from the nucleus of a neuron like the roots of a tree.
Dendrites receive signals from other neurons through synapses, passing them from the tip to the barrel to the core. According to a group of researchers from Stanford University, in computing technology “nanodendrites” could function similarly.
Cooperating with the semiconductor manufacturer Globalfoundries, researchers showed one such nanodendrite at the aforementioned IEEE conference. The device, which is a modified transistor, acts as a switch that recognizes the sequence of pulses of voltage in the length of a microsecond. It passes the current only with the correct order of these pulses, promising effective parallel processing in 3D chips, which are increasingly used in artificial intelligence.
Electrical Professor Philip Wong notes that heat is a “fundamental problem” of modern 3D chips. By emulating the dendrites of the brain, such chips will consume less energy and, importantly, release much less heat.
The device retains three main elements of the transistor: the source, shutter, and stock. The key difference is that the shutter of the device is divided into three parts and contains a thin layer of segnelectric material, allowing it to switch polarization when exposed to an electric field.
To move the charge along the transistor channel, a series of voltage pulses in the correct order must be applied, starting from the site closest to the source.
After the first shutter section receives an impulse, the charge carriers flow from the source to this section, and its polarization changes. The next impulse does the same in the middle section, stretching the charge carriers from the first section. Finally, the third section receives an impulse, completing the conducting channel.
The current iteration of the device includes a three-section shutter, which is the simplest version of such a structure, but the Stanford team plans to increase its segmentation in the future.
In addition, the creation of such 3D devices requires the development of new production processes. For example, as Professor Wong noted, the specified chips must be made at a very low temperature, so establishing their effective mass production remains an urgent issue for