Neuromorphic devices with molecular semiconductors.

Since the construction of Von Neumann architecture, up to these days, computers have been used to facilitate our everyday life by storing information and executing numerical calculations, but most importantly to perform AI tasks. Over time, many circuits and algorithms have been constructed to optimize the performance of these devices, granting access to more sophisticated, complex tasks to be executed in a fast accurate manner. Nowadays, computers are very efficient in task execution; However, traditional computing architecture is reaching its limitation due to many fundamental problems, such as CMOS scalability limits(Moore’s law), and the huge energy consumption due to the continuous information flow and conversion between memory and processor. In contrast, nature provided us a very compact, energy-efficient biological memory and processor combined, the human brain. Functioning as a memory capable of self-learning and processing incoming information, the brain houses approximately 10^(11) neurons, interconnected with around 10^(14) synapses, contained within a volume of 140 x 167 x 93 mm3 and weighing an average of 1.3 Kg. The most important property of the brain, apart from the huge number of neurons, is the synaptic plasticity, enabling continuous information acquisition, modification or erase through timing modulation of presynaptic and postsynaptic action potentials, this action leads to a modification in the postsynaptic Ca^(2+) signal, resulting in long (short)-term potentiation or depression. Neuromorphic computing (NC) mimics brain performance, thus allow computing and storage in a single unit (IMC), resulting an extremely short latency, and low energy consumption. This approach is particularly useful in dynamic vision sensors in self driving cars, and event driven sensors in robotics.