Dataset Information

Tree-based machine learning performed in-memory with memristive analog CAM.

ABSTRACT: Tree-based machine learning techniques, such as Decision Trees and Random Forests, are top performers in several domains as they do well with limited training datasets and offer improved interpretability compared to Deep Neural Networks (DNN). However, these models are difficult to optimize for fast inference at scale without accuracy loss in von Neumann architectures due to non-uniform memory access patterns. Recently, we proposed a novel analog content addressable memory (CAM) based on emerging memristor devices for fast look-up table operations. Here, we propose for the first time to use the analog CAM as an in-memory computational primitive to accelerate tree-based model inference. We demonstrate an efficient mapping algorithm leveraging the new analog CAM capabilities such that each root to leaf path of a Decision Tree is programmed into a row. This new in-memory compute concept for enables few-cycle model inference, dramatically increasing 10³ × the throughput over conventional approaches.

SUBMITTER: Pedretti G

PROVIDER: S-EPMC8490381 | biostudies-literature | 2021 Oct

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Tree-based machine learning performed in-memory with memristive analog CAM.

Pedretti Giacomo G Graves Catherine E CE Serebryakov Sergey S Mao Ruibin R Sheng Xia X Foltin Martin M Li Can C Strachan John Paul JP

Nature communications 20211004 1

Tree-based machine learning techniques, such as Decision Trees and Random Forests, are top performers in several domains as they do well with limited training datasets and offer improved interpretability compared to Deep Neural Networks (DNN). However, these models are difficult to optimize for fast inference at scale without accuracy loss in von Neumann architectures due to non-uniform memory access patterns. Recently, we proposed a novel analog content addressable memory (CAM) based on emergin ...[more]

PMID: 34608133

Similar Datasets

Project description:In this work, the discovery of volatile memristive devices that exhibit analog resistive switching (RS) and synaptic emulation based on squaraine materials is presented. Specifically, organic microtubes (MTs) based on 2,4-bis[(4-(N,N-diisobutyl)-2-6-hydroxyphenyl]squaraine (SQ) are prepared by evaporation-induced self-assembly (EISA). The MTs are ca. 2 μm in diameter (aspect ratio: 10-130). While powder X-ray diffraction data for MTs identify monoclinic and orthorhombic polymorphs, optical data report the monoclinic phase with energetic disorder. By favorable energetic alignment of the Au work function with the SQ HOMO energy, unipolar (hole-only) symmetric metal-insulator-metal devices are formed by EISA of MT meshes on interdigitated electrodes. The DC I-V characteristics acquired exhibit pinched hysteretic I-V loops, indicative of memristive behavior. Analysis indicates Ohmic transport at low bias with carrier extraction by thermionic emission. At high bias, space-charge-limited conduction in the presence of traps distributed in energy, enhanced by a Poole-Frenkel effect and with carrier extraction by Fowler-Nordheim tunneling, is observed. These data indicate purely electronic conduction. I-V hysteresis attenuates at smaller voltage windows, suggesting that carrier trapping/detrapping underpins the hysteresis. By applying triangular voltage waveforms, device conductance gradually increases sweep-on-sweep, with wait-time-erase or voltage-erase options. Using square waveforms, repeated erase-write-read of multiple distinct conductance states is achieved. Such analog RS behavior is consistent with trap filling/emptying effects. By waveform design, volatile conductance states may also be written so that successive conductance states exhibit identical current levels, indicating forgetting of previously written states and mimicking the forgetting curve. Finally, advanced synaptic functions, i.e., excitatory postsynaptic current, paired-pulse facilitation, pulse-dependent plasticity, and a transition from short- to long-term memory driven by post-tetanic potentiation, are demonstrated.

Dataset Information

Tree-based machine learning performed in-memory with memristive analog CAM.

Publications

Tree-based machine learning performed in-memory with memristive analog CAM.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets