Nanosecond protonic programmable resistors for analog deep learning

Abstract

Nanoscale ionic programmable resistors for analog deep learning are 1000 times smaller than biological cells, but it is not yet clear how much faster they can be relative to neurons and synapses. Scaling analyses of ionic transport and charge-transfer reaction rates point to operation in the nonlinear regime, where extreme electric fields are present within the solid electrolyte and its interfaces. In this work, we generated silicon-compatible nanoscale protonic programmable resistors with highly desirable characteristics under extreme electric fields. This operation regime enabled controlled shuttling and intercalation of protons in nanoseconds at room temperature in an energy-efficient manner. The devices showed symmetric, linear, and reversible modulation characteristics with many conductance states covering a 20× dynamic range. Thus, the space-time-energy performance of the all–solid-state artificial synapses can greatly exceed that of their biological counterparts.

Molecularly Hybridized Conduction in DPP‐Based Donor–Acceptor Copolymers toward High‐Performance Iono‐Electronics

“…Ionic memory has recently garnered a great deal of attention, especially in neuromorphic computing. [9,57,58] Not only does the use of ion insertion in memory devices emulate the mechanism found in the biological memory system, but it has also been shown to be energy efficient, deployable to various classes of materials, and promising toward high-fidelity ANNs. [9,57] To test the utility of the ngT polymers in advanced iono-electronics, we fabricated electrochemical neuromorphic devices (ENDs) and characterized the impact of tuned gT percentage on the electrochemical synaptic plasticity.…”

“…[9,57,58] Not only does the use of ion insertion in memory devices emulate the mechanism found in the biological memory system, but it has also been shown to be energy efficient, deployable to various classes of materials, and promising toward high-fidelity ANNs. [9,57] To test the utility of the ngT polymers in advanced iono-electronics, we fabricated electrochemical neuromorphic devices (ENDs) and characterized the impact of tuned gT percentage on the electrochemical synaptic plasticity. In these devices, a gate electrode is used to mimic a pre-synaptic neuron and send electrical pulses to tune the injection and extraction of ions from the electrolyte to the semiconducting channel.…”

Liquid-Based Memory Devices for Next-Generation Computing

“…However, silicon-based memory devices cannot easily meet the needs of the future development of data-storage devices owing to their physical and technological limitations, such as difficulty in fabrication processes, high fabrication cost, and high power consumption. ,− Resistive-switching memory (RSM) devices have emerged as next-generation memory devices owing to their simple structure, good scalability, and the wide variety of materials that may be used in their development. ,, RSM devices typically have two-terminal metal/insulator/metal structures . The characteristics of RSM can be optimized for high performance by designing low operating energy, fast operating speed, and high stability, depending on the materials of the insulating layer. − In addition, RSM devices can emulate synaptic functions through the use of analog resistive-switching characteristics. , The emulation of synaptic functions has emerged because they improve the performance of memory devices. − Synaptic functions are highly efficient computing processes in the brain that can handle complex computations with extremely low energy consumption, and they are emulated using memory devices called neuromorphic devices . Neuromorphic computing emulates synaptic plasticity, backpropagation learning, and synaptic-weight updates in the brain. − Changing the strength of connections between neurons, synaptic plasticity, and synaptic-weight updates are assumed to be mechanisms of memorization and computation in the brain. − Synaptic plasticity induces memory formation and information storage, whereas backpropagation learning enables the computation of the gradient of an objective function with respect to synaptic weight for parallel computing , …”

Photonic tensor core for machine learning: a review

“…5 Analog electronic solutions have been proposed in the latest years to address these limitations, with FPGA solutions, 6 PCM, 7 memsistor, 8 and protonic approaches. 9 But the underlying limits of the electron to work at high speed in an energy-efficient way still stay. Optics and photonics, on the other hand, can overcome those limitations by relying on the electromagnetic behavior of the light to perform neural network tasks.…”