Search

Two-dimensional (2D) transition-metal dichalcogenides (TMDs) materials, such as molybdenum disulfide (MoS2), stand out due to their atomically thin layered structure and exceptional electrical properties. Consequently, they could potentially become one of the main materials for future integrated high-performance logic circuits. However, the local back-gate-based MoS2 transistors on a silicon substrate can lead to the degradation of electrical characteristics. This degradation is caused by the abnormal effect of gate sidewalls, leading to non-uniform field controllability. Therefore, the buried-gate-based MoS2 transistors where the gate electrodes are embedded into the silicon substrate are fabricated. The several device parameters such as field-effect mobility, on/off current ratio, and breakdown voltage of gate dielectric are dramatically enhanced by field-effect mobility (from 0.166 to 1.08 cm2/V·s), on/off current ratio (from 4.90 × 105 to 1.52 × 107), and breakdown voltage (from 15.73 to 27.48 V) compared with a local back-gate-based MoS2 transistor, respectively. Integrated logic circuits, including inverters, NAND, NOR, AND, and OR gates, were successfully fabricated by 2-inch wafer-scale through the integration of a buried-gate MoS2 transistor array.

The first-generation of BrainScaleS, also referred to as BrainScaleS-1, is a neuromorphic system for emulating large-scale networks of spiking neurons. Following a ‘physical modeling’ principle, its VLSI circuits are designed to emulate the dynamics of biological examples: analog circuits implement neurons and synapses with time constants that arise from their electronic components’ intrinsic properties. It operates in continuous time, with dynamics typically matching an acceleration factor of 10 000 compared to the biological regime. A fault-tolerant design allows it to achieve wafer-scale integration despite unavoidable analog variability and component failures. In this paper, we present the commissioning process of a BrainScaleS-1 wafer module, providing a short description of the system’s physical components, illustrating the steps taken during its assembly and the measures taken to operate it. Furthermore, we reflect on the system’s development process and the lessons learned to conclude with a demonstration of its functionality by emulating a wafer-scale synchronous firing chain, the largest spiking network emulation ran with analog components and individual synapses to date.

In this study, we propose and analyze in simulations a new, highly flexible method of imple-menting synaptic plasticity in a wafer-scale, accelerated neuromorphic hardware system. Thestudy focuses on globally modulated STDP, as a special use-case of this method. Flexibility isachieved by embedding a general-purpose processor dedicated to plasticity into the wafer. Toevaluate the suitability of the proposed system, we use a reward modulated STDP rule in a spiketrain learning task. A single layer of neurons is trained to fire at specific points in time withonly the reward as feedback. This model is simulated to measure its performance, i.e. the in-crease in received reward after learning. Using this performance as baseline, we then simulatethe model with various constraints imposed by the proposed implementation and compare theperformance. The simulated constraints include discretized synaptic weights, a restricted inter-face between analog synapses and embedded processor, and mismatch of analog circuits. Wefind that probabilistic updates can increase the performance of low-resolution weights, a simpleinterface between analog synapses and processor is sufficient for learning, and performance isinsensitive to mismatch. Further, we consider communication latency between wafer and theconventional control computer system that is simulating the environment. This latency increasesthe delay, with which the reward is sent to the embedded processor. Because of the time continu-ous operation of the analog synapses, delay can cause a deviation of the updates as compared tothe not delayed situation. We find that for highly accelerated systems latency has to be kept to aminimum. This study demonstrates the suitability of the proposed implementation to emulatethe selected reward modulated STDP learning rule. It is therefore an ideal candidate for imple-mentation in an upgraded version of the wafer-scale system developed within the BrainScaleSproject.