skip to main content


Search for: All records

Award ID contains: 1710009

Note: When clicking on a Digital Object Identifier (DOI) number, you will be taken to an external site maintained by the publisher. Some full text articles may not yet be available without a charge during the embargo (administrative interval).
What is a DOI Number?

Some links on this page may take you to non-federal websites. Their policies may differ from this site.

  1. In this paper, we propose a Flexible processing-in-DRAM framework named FlexiDRAM that supports the efficient implementation of complex bulk bitwise operations. This framework is developed on top of a new reconfigurable in-DRAM accelerator that leverages the analog operation of DRAM sub-arrays and elevates it to implement XOR2-MAJ3 operations between operands stored in the same bit-line. FlexiDRAM first generates an efficient XOR-MAJ representation of the desired logic and then appropriately allocates DRAM rows to the operands to execute any in-DRAM computation. We develop ISA and software support required to compute in-DRAM operation. FlexiDRAM transforms current memory architecture to a massively parallel computational unit and can be leveraged to significantly reduce the latency and energy consumption of complex workloads. Our extensive circuit-to-architecture simulation results show that averaged across two well-known deep learning workloads, FlexiDRAM achieves ∼15× energy-saving and 13× speedup over the GPU outperforming recent processing-in-DRAM platforms. 
    more » « less
  2. This work investigates the role of extra oxygen vacancies, introduced by a hydrogen plasma at midpoint of deposition of a 6 nm thick HfO2 to reduce the switching power consumption in a RRAM device. Initially TiN, which is a commonly used metal in CMOS technology, was used as the top electrode for treated HfO2. Subsequently Ru and TaN as top electrodes were explored to enhance the switching behavior and power consumption. A range of compliance currents from 1 nA to 1 µA were used to evaluate the switching characteristics. The role of both TaN and Ru as bottom metal was also evaluated. With Ru as top metal the device switched at a compliance current of 1 nA and higher. Whereas when Ru was used as bottom electrode, devices were unable to switch below a compliance current of 50 µA. For TaN as top metal electrode, devices switched at and above 1 µA CC whereas with TaN as bottom metal the initial switching was at CC of 2 µA. It was observed that use of Ru as a top metal significantly reduced the switching energy of the plasma treated HfO2 RRAM device but was ineffective when used as a bottom metal. 
    more » « less
  3. Artificial Neural Networks (ANNs) are currently being used as function approximators in many state-of-the-art Reinforcement Learning (RL) algorithms. Spiking Neural Networks (SNNs) have been shown to drastically reduce the energy consumption of ANNs by encoding information in sparse temporal binary spike streams, hence emulating the communication mechanism of biological neurons. Due to their low energy consumption, SNNs are considered to be important candidates as co-processors to be implemented in mobile devices. In this work, the use of SNNs as stochastic policies is explored under an energy-efficient first-to-spike action rule, whereby the action taken by the RL agent is determined by the occurrence of the first spike among the output neurons. A policy gradient-based algorithm is derived considering a Generalized Linear Model (GLM) for spiking neurons. Experimental results demonstrate the capability of online trained SNNs as stochastic policies to gracefully trade energy consumption, as measured by the number of spikes, and control performance. Significant gains are shown as compared to the standard approach of converting an offline trained ANN into an SNN. 
    more » « less
  4. Computing systems inspired by the architecture of the human brain is poised to revolutionize the engines for information processing and data analytics. However, the efficiency and performance of these platforms pale in comparison with the human brain, especially when benchmarked in terms of metrics such as intelligence per Watt per square mm. In this paper, we review some recent progress and future prospects of building artificial intelligence systems that target the efficiency of the brain, leveraging the unique properties of nanoscale memristive device technologies. 
    more » « less