The roadmap is organized into several thematic sections, outlining current computing challenges, discussing the neuromorphic computing approach, analyzing mature and currently utilized technologies, providing an overview of emerging technologies, addressing material challenges, exploring novel computing concepts, and finally examining the maturity level of emerging technologies while determining the next essential steps for their advancement.

In memristors and resistance switching devices, there is a region prior to switching which exhibits current transients with potentially useful dynamics. We refer to this region as the subthreshold region owing to it occurring prior to any switching threshold. These transients exhibit a characteristic peaked response with a fast rise in current followed by a slower decay. This behaviour has previously been used to quantify the mobilities of defects drifting within the active layer of the devices, but it has also been used in neuromorphic circuits to carry out edge detection, to implement homeostasis within artificial synapses and could have uses in replicating eligibility traces. We present an empirical SPICE model to reproduce these transients within circuit simulators. The model is compared with experimental datasets for a range of applied voltages and we present experimentally verified parameters for readers to use within their own simulations.

Digital computers have been getting exponentially faster for decades, but huge challenges exist today. Transistor scaling, described by Moore's law, has been slowing down over the last few years, ending the era of fully predictable performance improvements. Furthermore, the data-centric computing demands fueled by machine learning applications are rapidly growing, and current computing systems -- even with the historical rate of improvements driven by Moore's law -- cannot keep up with these enormous computational demands. Some are turning to analogue in-memory computing as a solution, where specialised systems operating on physical principles accelerate specific tasks. We explore how emerging nonvolatile memories can be used to implement such systems tailored for machine learning. In particular, we discuss how memristive crossbar arrays can accelerate key linear algebra operations used in neural networks, what technological challenges remain, and how they can be overcome.

A design concept of phase-separated amorphous nanocomposite thin films is presented that realizes interfacial resistive switching (RS) in hafnium-oxide-based devices. The films are formed by incorporating an average of 7% Ba into hafnium oxide during pulsed laser deposition at temperatures ≤400°C. The added Ba prevents the films from crystallizing and leads to ∼20-nm-thin films consisting of an amorphous HfOx host matrix interspersed with ∼2-nm-wide, ∼5-to-10-nm-pitch Ba-rich amorphous nanocolumns penetrating approximately two-thirds through the films. This restricts the RS to an interfacial Schottky-like energy barrier whose magnitude is tuned by ionic migration under an applied electric field. Resulting devices achieve stable cycle-to-cycle, device-to-device, and sample-to-sample reproducibility with a measured switching endurance of ≥104 cycles for a memory window ≥10 at switching voltages of ±2 V. Each device can be set to multiple intermediate resistance states, which enables synaptic spike-timing–dependent plasticity. The presented concept unlocks additional design variables for RS devices.

We present a resistance switching device that exhibits analogue potentiation and depression of conductance under the same voltage polarity. This contrasts with previously studied devices that potentiate and depress under opposite polarities. We refer to this mode of operation as the subthreshold regime due to it occurring at voltage or current biases that are insufficient to produce discrete or non-volatile switching. This behaviour has the potential to reduce the complexity of neuronal and synaptic circuitry in neuromorphic computing by removing the need for voltage pulses of both positive and negative polarities. The characteristically long timescales may also help replicate bio-realistic timings. In this article, we detail how to induce this unique behaviour, how to tune its properties to a desired response, and finally, we demonstrate one potential application.

In this study we present a resistance switching device that exhibits analogue potentiation and depression of conductance in the same polarity. This is in contrast to devices studied previously which exhibit potentiation and depression in opposite polarities. This has the potential to reduce complexity of the surrounding circuitry in neuromorphic computing by only requiring voltage pulses of a single polarity. In this paper, we detail how to induce this unique behaviour in devices as well as how to tune its properties to a desired response.

Filamentary resistance switching, or ReRAM, devices based on oxides suffer from device-do-device and cycle-to-cycle variability of electrical characteristics (electroforming voltages, set and reset voltages, resistance levels and cycling endurance). These are largely materials issues related to the microstructure of the switching oxide. Here we outline strategies to engineer the electrical performance of silicon oxide ReRAM by controlling the oxide microstructure at the nanometre scale through approaches including engineered interfaces and ion implantation. We demonstrate control over the distribution of switching voltages, electroforming voltages, and stable multilevel resistance states.

We report that implanting argon ions into a film of uniform atomic layer deposition (ALD)-grown SiOx enables electroforming and switching within films that previously failed to electroform at voltages <15 V. We note an implantation dose dependence of electroforming success rate: electroforming can be eliminated when the dosage is high enough. Our devices are capable of multi-level switching during both set and reset operations, and multiple resistance states can be retained for more than 30,000 s under ambient conditions. High endurance of more than 7 million (7.9 × 106) cycles is achieved alongside low switching voltages (±1 V). Comparing SiOx fabricated by this approach with sputtered SiOx we find similar conduction mechanisms between the two materials. Our results show that intrinsic SiOx switching can be achieved with defects created solely by argon bombardment; in contrast to defects generated during deposition, implantation generated defects are potentially more controllable. In the future, noble ion implantation into silicon oxide may allow optimization of already excellent resistance switching devices.

This study aims to contribute to the burgeoning field of brain‐inspired computing by expanding it beyond conventional fabrication methods. Herein, the obstacles toward the effective inkjet printing process are encountered and the electrical characteristics are explored, providing new insights into the reliability aspects of fully printed Ag/a‐TiO2/Ag electronic synapses. The versatility of the approach is further enhanced by the highly stable in‐house‐developed a‐TiO2 ink, exhibiting optimal shelf life of five months and repeatable jetting, producing layers with nanoscale thickness resolution. Most importantly, device electrical characterization reveals synaptic dynamics, leading to activity‐dependent conductance state retention and adaptation characteristics, implying inherent learning capabilities. The synaptic dynamics are attained by solely adjusting the duty cycle of the applied pulsed voltage trigger, while keeping amplitude and polarity fixed, a method readily compatible with realistic applications. Furthermore, I–V analysis demonstrates a dynamic range dependence on a‐TiO2 layer thickness and conduction mechanism that is akin to the conventionally developed electronic TiO2 synapses. The developed devices provide a time‐ and cost‐effective ecologically benign alternative toward biomimetic signal processing for future flexible neural networks.

In a data‐driven economy, virtually all industries benefit from advances in information technology—powerful computing systems are critically important for rapid technological progress. However, this progress might be at risk of slowing down if the discrepancy between the current computing power demands and what the existing technologies can offer is not addressed. Key limitations to improving energy efficiency are the excessive growth of data transfer costs associated with the von Neumann architecture and the fundamental limits of complementary metal–oxide–semiconductor (CMOS) technologies, such as transistors. Herein, three approaches that will likely play an essential role in future computing systems are discussed: memristive electronics, spintronics, and electronics based on 2D materials. The authors present how these technologies may transform conventional digital computers and contribute to the adoption of new paradigms, like neuromorphic computing.

The computing power demands to run artificial neural networks (ANNs) are increasing at rates much greater than improvements made with current CMOS-based technologies. The demand has contributed to a need for novel paradigms, including memristor-based accelerators. This work explores two algorithmic approaches to mitigate non-idealities inherent in most memristor-based systems. The first is to apply a concept of committee machines during inference, and the second is nonideality-aware training of memristor-based ANNs.

In a data-driven economy, virtually all industries benefit from advances in information technology—powerful computing systems are critically important for rapid technological progress. However, this progress might be at risk of slowing down if we do not address the discrepancy between our current computing power demands and what the existing technologies can offer. Key limitations to improving the energy efficiency are the excessive growth of data transfer costs associated with the von Neumann architecture and the fundamental limits of complementary metal–oxide–semiconductor (CMOS) technologies, such as transistors. In this perspective article, we discuss three technologies that will likely play an essential role in future computing systems: memristive electronics, spintronics, and electronics based on 2D materials. We present some representative applications for each and speculate how these could fit within future digital, quantum and neuromorphic systems.