Saptarshi Mandal

Novel synaptic memory device for neuromorphic computing

This report discusses the electrical characteristics of two-terminal synaptic memory devices capable of demonstrating an analog change in conductance in response to the varying amplitude and pulse-width of the applied signal. The devices are based on Mn doped HfO2 material. The mechanism behind reconfiguration was studied and a unified model is presented to explain the underlying device physics. The model was then utilized to show the application of these devices in speech recognition. A comparison between a 20 nm × 20 nm sized synaptic memory device with that of a state-of-the-art VLSI SRAM synapse showed ~10× reduction in area and >106 times reduction in the power consumption per learning cycle.

Switching dynamics and charge transport studies of resistive random access memory devices

We report the switching dynamics and charge transport studies on Ru/HfO2/TiOx/Ru resistive random access memory devices in low resistance state (LRS), high resistance state (HRS), and virgin resistance state (VRS). The charge transport in LRS is governed by Ohmic conduction of electrons through local filamentary paths while it is governed by a combination of Frenkel-Poole emission and trap assisted tunneling process in HRS and VRS. The area of the filament in LRS is extracted and related to the compliance current. The thickness of the re-oxidized filament is extracted and related to the reset voltage in HRS. The energy consumed during the reset process was analyzed on the time-scale to experimentally demonstrate joule-heating mediated oxidation dynamics of filament during device reset.

Effects of Mg-Doping on

We report the switching characteristics of Mg-doped HfO2-based ReRAM devices consisting of Ru/ Mg:HfO2/TiN/W stacks. The concentration of the Mg dopant was varied from 0% to 20% for four samples to show the impact on the forming voltage. In addition to reducing the forming voltage from 2.8 to 1.7 V, Mg doping in HfO2 also improved the repeatability in the initial device characteristics, switching characteristics, and retention. The mechanism of conduction was identified as Frenkel-Poole emission in doped and undoped samples in virgin resistance state (VRS). Further analysis showed that the increased conductance in the doped samples in VRS was due to higher carrier concentration.

{\rm HfO}_{2}

In this work we introduce a two-terminal memristive device using Mn doped HfO2. The devices can emulate synaptic behavior based on their transient characteristics. These properties can be exploited to show spike-timing based learning in a network of neurons and synapses. We use the device characteristics to simulate a 4×4 crossbar array of synapses and observe the evolution of the weights over time. The effect of device variability on the performance of synaptic weight update has been examined based on different test conditions of initial randomness and variation in percentage change of strength during spike-timing based updates. Some inferences have been drawn regarding the need of additional circuits for improving reliability of the cross-bar arrays. We believe this study is critical in assessing the design constraints and requirements necessary for integrating memristive devices in crossbars for spike based computations.

-Based ReRAM Device Switching Characteristics

This paper discusses the impact of the bidirectional diode parameters on the read failures in 1ReRAM 1Diode (1D1R) crossbar array memory architectures. Our studies show that while a diode is integral for the successful read operation, the maximum achievable crossbar memory capacity is a strong function of the reverse saturation current of the diode. An acceptable reverse saturation current target for diodes should be 0.1 A/cm2 for 10 nm × 10 nm ReRAM cell for high density memory. For multi-level-cell (MLC) operation, read failure for multiple high resistance states is limited by the reverse saturation current of diode while the line resistance of crossbar arrays plays significant role for read failure of multiple low resistance states.

Doped HfO 2 based nanoelectronic memristive devices for self-learning neural circuits and architecture

We report the synaptic characteristics of novel two-terminal reconfigurable devices fabricated using a doped transition metal oxide. These devices demonstrate short-term plasticity, frequency-dependent synaptic augmentation, long-term potentiation, and long-term depression, and have a potential to show spike timing-dependent plasticity that are macroscopically similar to a biological synapse. The underlying mechanism behind the observed synaptic characteristics was studied using charge transport characterization. Based on this study, a fundamental correlation between the governing device physics and the synaptic characteristic has been established. We believe that by carefully engineering the dopants, the synaptic transmission of these devices can be modulated, which will provide a viable route to replicate the functional diversity of a biological neural system on chip.

Understanding the impact of diode parameters on sneak current in 1Diode 1ReRAM crossbar architectures

To realize extreme-scale neuromorphic computation inspired by a biological brain, there is a need to develop two-terminal reconfigurable devices that can mimic the low-power specifications and scalability of a biological synapse. This paper discusses the synaptic characteristics of doped transition metal oxide based two-terminal devices. Spike-frequency dependent augmentation in conductance was observed. In addition, the devices could be reconfigured to different conductance states by changing the input pulse-width. This characteristic was used to demonstrate spike-timing dependent plasticity (STDP). The mechanism of reconfiguration is also briefly discussed.

Study of Synaptic Behavior in Doped Transition Metal Oxide-Based Reconfigurable Devices

We report an experimental study to understand the reduction in the forming voltage with slow electro-forming of HfO2 based ReRAM devices. Using a combination of capacitance-voltage and current-voltage measurements, we captured the change in capacitance due to dielectric polarization as function of voltage sweep rates. The dielectric polarization was attributed to the charge trapping or internal redistribution of charged centers under electric field. Retention testing showed that this change in capacitance was volatile and decayed to initial values over time after the removal of bias indicating a dielectric relaxation. The dielectric polarization was significantly higher when voltage-sweep rate was slow which causes electro-forming of the dielectric at lower forming voltages.