Resistive in-memory computing (IMC) architectures currently lag behind SRAM IMCs and digital accelerators in both energy efficiency and compute density due to their low compute accuracy. This article proposes the use of signal-to-noise-plus-distortion ratio (SNDR) to quantify the compute accuracy of IMCs and identify the device, circuit, and architectural parameters that affect it. We further analyze the fundamental limits on the SNDR of magnetoresistive random access memory (MRAM-), resistive random access memory (ReRAM-), and ferroelectric field effect transistor (FeFET)-based IMCs employing parameter variation and noise models that were validated against measured results from a recent MRAM-based IMC prototype in a 22 nm process. At high-output signal magnitude, we can find that the maximum achievable SNDR is limited by the pre-analog-to-digital-converter (ADC) array nonidealities, such as the conductance variations (CVs), parasitic resistances, and current mirror mismatch (MM), whereas the ADC thermal (AT) noise limits the SNDR at small signal magnitudes. Furthermore, for large dot-product (DP) dimensions ( $N > 50$ ), the maximum achievable SNDR is highest for FeFET, followed by ReRAM and then MRAM. Finally, the increase in conductance contrast ( ${g_ {text {ON}} }/ {g_ {text {OFF}} }$ ) enhances the maximum achievable SNDR only until it reaches a value of approximately 12. ReRAMs and FeFETs demonstrate high energy efficiencies while achieving high SNDR, as their low conductance values lead to lower currents and lower noise due to wire parasitics. In all cases, across all three device types, DP dimension, ADC precision, and conductance contrast, the maximum achievable SNDR is found to be in the range of 18–22 dB, barely meeting the minimum needed for achieving an inference accuracy close to an equivalent fixed-point digital architecture. Finally, we demonstrate a network-level accuracy of 84.5% when mapping an ResNet-20 (CIFAR-10) by ReRAM-based architecture at a SNDR of 22 dB, in which MRAM- and FeFET-based architectures cannot realize. This result clearly implies the need for other approaches, e.g., algorithmic- and learning-based methods, to improve the inference accuracy of resistive IMC architectures.Resistive in-memory computing (IMC) architectures currently lag behind SRAM IMCs and digital accelerators in both energy efficiency and compute density due to their low compute accuracy. This article proposes the use of signal-to-noise-plus-distortion ratio (SNDR) to quantify the compute accuracy of IMCs and identify the device, circuit, and architectural parameters that affect it. We further analyze the fundamental limits on the SNDR of magnetoresistive random access memory (MRAM-), resistive random access memory (ReRAM-), and ferroelectric field effect transistor (FeFET)-based IMCs employing parameter variation and noise models that were validated against measured results from a recent MRAM-based IMC prototype in a 22 nm process. At high-output signal magnitude, we can find that the maximum achievable SNDR is limited by the pre-analog-to-digital-converter (ADC) array nonidealities, such as the conductance variations (CVs), parasitic resistances, and current mirror mismatch (MM), whereas the ADC thermal (AT) noise limits the SNDR at small signal magnitudes. Furthermore, for large dot-product (DP) dimensions ( $N > 50$ ), the maximum achievable SNDR is highest for FeFET, followed by ReRAM and then MRAM. Finally, the increase in conductance contrast ( ${g_ {text {ON}} }/ {g_ {text {OFF}} }$ ) enhances the maximum achievable SNDR only until it reaches a value of approximately 12. ReRAMs and FeFETs demonstrate high energy efficiencies while achieving high SNDR, as their low conductance values lead to lower currents and lower noise due to wire parasitics. In all cases, across all three device types, DP dimension, ADC precision, and conductance contrast, the maximum achievable SNDR is found to be in the range of 18–22 dB, barely meeting the minimum needed for achieving an inference accuracy close to an equivalent fixed-point digital architecture. Finally, we demonstrate a network-level accuracy of 84.5% when mapping an ResNet-20 (CIFAR-10) by ReRAM-based architecture at a SNDR of 22 dB, in which MRAM- and FeFET-based architectures cannot realize. This result clearly implies the need for other approaches, e.g., algorithmic- and learning-based methods, to improve the inference accuracy of resistive IMC architectures. Leer más
Related posts:
PoE-Enabled Visible Light Positioning Network With Low Bandwidth Requirement and High Precision Pulse Reconstruction
Self-Organizing Mapping Neural Network Implementation Based on 3-D NAND Flash for Competitive Learning
Secure Initial Access and Beam Alignment Using Deep Learning in 5G and Beyond Systems
Impact of Technology Scaling and Back-End-of-the-Line Technology Solutions on Magnetic Random-Access Memories
Improved Forest Signal Detection for Space-Borne Photon-Counting LiDAR Using Automatic Machine Learning
Cam-PC: A Novel Method for Camouflaging Point Clouds to Counter Adversarial Deception in Remote Sensing