Digital computing is nearing its physical limits as computing needs and energy consumption rapidly increase. Analogue‐memory‐based neuromorphic computing can be orders of magnitude more energy efficient at data‐intensive tasks like deep neural networks, but has been limited by the inaccurate and unpredictable switching of analogue resistive memory. Filamentary resistive random access memory (RRAM) suffers from stochastic switching due to the random kinetic motion of discrete defects in the nanometer‐sized filament. In this work, this stochasticity is overcome by incorporating a solid electrolyte interlayer, in this case, yttria‐stabilized zirconia (YSZ), toward eliminating filaments. Filament‐free, bulk‐RRAM cells instead store analogue states using the bulk point defect concentration, yielding predictable switching because the statistical ensemble behavior of oxygen vacancy defects is deterministic even when individual defects are stochastic. Both experiments and modeling show bulk‐RRAM devices using TiO2‐X switching layers and YSZ electrolytes yield deterministic and linear analogue switching for efficient inference and training. Bulk‐RRAM solves many outstanding issues with memristor unpredictability that have inhibited commercialization, and can, therefore, enable unprecedented new applications for energy‐efficient neuromorphic computing. Beyond RRAM, this work shows how harnessing bulk point defects in ionic materials can be used to engineer deterministic nanoelectronic materials and devices.

No abstract

The biological brain has set a golden standard in computational efficiency, both due to its massive parallelism, and its ability to perform in-memory computing within the same ionic substrate. Ion-gated channels determine synaptic strength which is known to be a mechanism for memory storage, and the very same ions that pass through these gates encode data in the form of spikes. Communication, computation, and storage all occur within the same local medium. The brain's ability to perform in-memory processing within a unified ionic mechanism has driven many researchers to apply ion-driven non-volatile memories to emulate learning rules at the device-level. [1-12] The operating principles of resistive random access memory (RRAM) draw parallel with biological synapses. From a physical standpoint, the top electrode (TE) corresponds to a pre-synaptic terminal, the insulator layer acts as the substrate through which neurotransmitters are released, and the bottom electrode (BE) Biologically plausible computing systems require fine-grain tuning of analog synaptic characteristics. In this study, lithium-doped silicate resistive random access memory with a titanium nitride (TiN) electrode mimicking biological synapses is demonstrated. Biological plausibility of this RRAM device is thought to occur due to the low ionization energy of lithium ions, which enables controllable forming and filamentary retraction spontaneously or under an applied voltage. The TiN electrode can effectively store lithium ions, a principle widely adopted from battery construction, and allows state-dependent decay to be reliably achieved. As a result, this device offers multi-bit functionality and synaptic plasticity for simulating various strengths in neuronal connections. Both short-term memory and long-term memory are emulated across dynamical timescales. Spike-timing-dependent plasticity and paired-pulse facilitation are also demonstrated. These mechanisms are capable of self-pruning to generate efficient neural networks. Time-dependent resistance decay is observed for different conductance values, which mimics both biological and artificial memory pruning and conforms to the trend of the biological brain that prunes weak synaptic connections. By faithfully emulating learning rules that exist in human's higher cortical areas from STDP to synaptic pruning, the device has the capacity to drive forward the development of highly efficient neuromorphic computing systems.

Wireless high-speed communication standards such as WiFi, WiMax and LTE use spectrally-efficient OFDM modulation that encodes signal information in both amplitude and phase. Use of this non-constant envelope modulation requires a linear PA, operating at a less-than-peak signal level to realize higher linearity and inherently reduced efficiency. Because the PA is the dominant power consumer in most RF transceivers, operation with reduced efficiency leads to short battery lifetime and reduced mobility. Consequently, many efforts to utilize more efficient switching amplifiers with linearization circuitry have been made, notably through pulse-width modulation [1], outphasing [2] and envelope elimination and restoration (EER) [3,4]. Of the three, EER offers the best performance tradeoff between linearity, output power and efficiency; however, most previous implementations have come at the cost of large, power-hungry analog supply modulators. Additionally, conventional EER techniques are subject to nonlinearity induced by delay mismatch between the amplitude-and phasemodulated signal components. An alternative solution modulated the output power by selecting multiple PA unit cells [5], but this exhibits low efficiency at low output power levels because the power control is achieved by changing the total PA transconductance through switching of inefficient cells.This paper introduces an EER 90nm CMOS experimental prototype switchedcapacitor power amplifier (SCPA) that achieves high output power, efficiency and linear output-power control using a switched-capacitor-based switching PA without the use of a supply modulator. While amplifying 64-QAM OFDM modulation with a 20MHz signal bandwidth it achieves an average output power of 17.7dBm, an average PAE of 32.1%, and an EVM of 2.9%.Switched-capacitor circuit techniques are widely used in analog/mixed-signal design because capacitors are area-efficient native devices and CMOS transistors are excellent switches [6]. High-accuracy capacitor ratios coupled with digital signal processing techniques are easily applied to switched-capacitor circuits. These techniques can now be adopted directly at RF frequencies because of the higher operating speeds with scaled CMOS. In a switched-capacitor circuit, any voltage can be generated based on the ratio of the capacitors switched to V DD or ground (V GND ). It is important to note that there is no loss of energy ideally in the charge redistribution among capacitors ( Fig. 24.3.1). To efficiently generate a desired output voltage, capacitors are selectively connected to either V GND or switched between V GND and V DD . Hence, the ratio of capacitors switching (ΣC on ) between V GND and V DD compared to the total capacitance (ΣC on +ΣC off ) defines the output voltage. The capacitors are switched at the desired RF carrier frequency; a bandpass filter (BPF) (e.g., matching network) is created by connecting an inductive reactance in series with the capacitor array to select the RF signal to be broadcast by the SCPA. Because V DD and V GND...

No abstract