Tailor-made synaptic dynamics based on memristive devices

Abstract

The proliferation of machine learning algorithms in everyday applications such as image recognition or language translation has increased the pressure to adapt underlying computing architectures towards these algorithms. Application specific integrated circuits (ASICs) such as the Tensor Processing Units by Google, Hanguang by Alibaba or Inferentia by Amazon Web Services were designed specifically for machine learning algorithms and have been able to outperform CPU based solutions by great margins during training and inference. As newer generations of chips allow handling of and computation on more and more data, the size of neural networks has dramatically increased, while the challenges they are trying to solve have become more complex. Neuromorphic computing tries to take inspiration from biological information processing systems, aiming to further improve the efficiency with which these networks can be trained or the inference can be performed. Enhancing neuromorphic computing architectures with memristive devices as non-volatile storage elements could potentially allow for even higher energy efficiencies. Their ability to mimic synaptic plasticity dynamics brings neuromorphic architectures closer to the biological role models. So far, memristive devices are mainly investigated for the emulation of the weights of neural networks during training and inference as their non-volatility would enable both processes in the same location without data transfer. In this paper, we explore realisations of different synapses build from memristive ReRAM devices, based on the Valence Change Mechanism. These synapses are the 1R synapse, the NR synapse and the 1T1R synapse. For the 1R synapse, we propose three dynamical regimes and explore their performance through different synapse criteria. For the NR synapse, we discuss how the same dynamical regimes can be addressed in a more reliable way. We also show experimental results measured on ZrOx devices to support our simulation based claims. For the 1T1R synapse, we explore the trade offs between the connection direction of the ReRAM device and the transistor. For all three synapse concepts we discuss the impact of device-to-device and cycle-to-cycle variability. Additionally, the impact of the stimulation mode on the observed behavior is discussed.

Effect of Transistor Transfer Characteristics on the Programming Process in 1T1R Configuration

“…In this configuration, the control of the channel via the gate voltage is weakened because the voltage at both the drain and the source node is nonzero. In other words, the transistor exhibits the body effect and its transfer characteristics become more complex [19]. In this article, only the calculation of the intrinsic voltage of the VCM cell during the SET process is discussed.…”

“…The intrinsic voltage of the VCM cell in the 1T1R structure can be calculated based on the transistor load line concept, as shown in Fig. 1(b) [19], [20]. Considering the VCM cell as the load of the transistor, the intersection of the transistor transfer characteristics and the load line is the operating current of the 1T1R structure.…”

“…For the simulations, we used the physics-based JART VCM v1b compact model which was shown to be able to describe various filamentary VCM systems like ZrO x and HfO x [15], [19], [22]. The device stack structure is approximated by the model structure.…”

Improved Arithmetic Performance by Combining Stateful and Non‐Stateful Logic in Resistive Random Access Memory 1T–1R Crossbars

“…It has to be noticed that in any 1T–1R structure which features bipolar memory devices, e.g., ReRAM, the body effect in the MOSFET transistor has to be considered. [ 45 ] This effect will lead to a shift in the transistor threshold voltage and a modified current conduction through the transistor depending on the applied bulk‐source voltage, . In the MAGIC operation the body effect marks one of several error sources which are discussed in more detail in Section 3.3.…”

From Oxides to 2D Materials: Advancing Memristor Technologies for Energy-Efficient Neuromorphic Computing

“…The analog switching characteristics of the emerging devices play a pivotal role in neuromorphic computing. − These characteristics enable the replication of synaptic connections in a manner that closely resembles that of biological synapses in the human brain. Unlike traditional computing architectures, which rely on digital representations and discrete state changes, neuromorphic computing capitalizes on the continuous and analog nature of these emerging devices to model complex interactions more accurately within neural networks. − Furthermore, the analog properties of these devices enable continuous weight changes in neural networks.…”