Stochastic neurons based on CuS/GeSe conductive bridge threshold switching memristors are designed to mimic the probabilistic computing ability of the brain at a hardware level.

In this letter, we fabricated a C-doped GeTe _x Ovonic threshold switching (OTS) selector in a via-hole structure. The material composition is simple without toxic element doping. Performance of the GeTe _x -based selector is improved to be comparable to commercial GeSe-based selectors through C doping. A selector with advantages of low off-current (2 nA), low threshold voltage (1.26 V) accompanying with a big voltage window (0.55 V), high selectivity ( > 4.2×10 ⁴ ), satisfactory endurance (> 10 ⁷ cycles with 2 mA on-current), low cycle-to-cycle V _th variation and fast operating speed (8.5 ns) was fabricated. Besides, selector with self-limited on-current character was found. Next, reasons behind off-current reduction and consistency improvement by using C doping were studied by First-principles calculations.

We demonstrate a CuS/GeSe dual-layer selector with excellent performance. This CuS/GeSe dual-layer selector exhibits ultrahigh selectivity (1.25 × 10 ⁹ ), high on-current drive (600 μA), ultralow off current (~100 fA), and extremely steep switching slope (<;1.4 mV/dec). This selector shows significant threshold switching behavior by introducing CuS as the ion supply layer to implant limited Cu ions into GeSe layer, to form unstable filaments at high voltage and retract Cu ions at low voltage. In addition, Cu/GeSe devices were fabricated as a comparison, showing desirable memory switching. These two devices are of great potential for cross-point memory applications.

The nanosize confined effect is believed to contribute to improving the resistance drift in nanophase change devices. However, the smaller dimension of device designs is limited by plane lithography techniques. Phase change memory with a confined thickness of ultrathin GeTe layers is fabricated to overcome the limit of current plane lithography. Those memory cells composed of two-dimensional materials present a suppressed resistance drift in their amorphous phase. The drift exponent ν is reduced to 0.05 for 3 nm GeTe layers. Combined with Raman spectroscopy and ab initio molecular dynamics simulations, the structural relaxation process is described as the decay of tetrahedral-bonded sites. Tetrahedrons in ultrathin films are more stable than those in bulk materials. The local motifs of amorphous GeTe ultrathin films are covalently bonded and highly ordered in a short range. The majority of highly ordered tetrahedral clusters prevents spontaneous structural relaxation and leads to high stability in amorphous states, which helps to stop intrinsic fluctuations in physical properties of SET and RESET states, without an extra processing cost.

Phase change materials, which has been focused on the non-volatile memory field, show the possibility to carry out data storage and computing in the same physical location. However, the resistance drift behavior of phase change memory has been a huge barrier not only to traditional binary memory application for a long time, but to multi-level storage and therefore the neural network computing. Here, a bipolar programming scheme is exploited to achieve drift-reduced intermediate states and convolutional neural network (CNN) computations in Ge ₂ Sb ₂ Te ₅ (GST) based memory cells. Experiments show that the resistance drift phenomena under bipolar programming have been reduced. Furthermore, the impact of bipolar operation on CNN for inference is investigated. This work provides effective means for implementing phase change neuromorphic processor with enhanced stability.