This paper explores the scalability of in-plane and perpendicular MTJ based STT-MRAMs from 65 nm to 8 nm while taking into consideration realistic variability effects. We focus on the read and write performances of a STT-MRAM based cache rather than the obvious advantages such as the denser bit-cell and zero static power. An accurate MTJ macromodel capturing key MTJ properties was adopted for efficient Monte Carlo simulations. For the simulation of access devices and peripheral circuitries, ITRS projected transistor parameters were utilized and calibrated using the MASTAR tool that has been widely used in industry. 6T SRAM and STT-MRAM arrays were implemented with aggressive assist schemes to mimic industrial memory designs. A constant JC0·RA/VDD scaling scenario was used which to the first order gives the optimal balance between read and write margins of STT-MRAMs. The thermal stability factor ensuring a 10 year retention time was obtained by adjusting the free layer thickness as well as assuming improvement in the crystalline anisotropy. Our studies based on the proposed scaling methodology show that in-plane STT-MRAM will outperform SRAM from 15 nm node, while its perpendicular counterpart requires further innovations in MTJ material in order to overcome the poor write performance scaling from 22 nm node onwards.

This article (Colonnade) presents a fully digital bit-serial compute-in-memory (CIM) macro. The digital CIM macro is designed for processing neural networks with reconfigurable 1-16 bit input and weight precisions based on bit-serial computing architecture and a novel all-digital bitcell structure. A column of bitcells forms a column MAC and used for computing a multiply-and-accumulate (MAC) operation. The column MACs placed in a row work as a single neuron and computes a dot-product, which is an essential building block of neural network accelerators. Several key features differentiate the proposed Colonnade architecture from the existing analog and digital implementations. First, its full-digital circuit implementation is free from process variation, noise susceptibility, and data-conversion overhead that are prevalent in prior analog CIM macros. A bitwise MAC operation in a bitcell is performed in the digital domain using a custom-designed XNOR gate and a full-adder. Second, the proposed CIM macro is fully reconfigurable in both weight and input precision from 1 to 16 bit. So far, most of the analog macros were used for processing quantized neural networks with very low input/weight precisions, mainly due to a memory density issue. Recent digital accelerators have implemented reconfigurable precisions, but they are inferior in energy efficiency due to significant off-chip memory access. We present a regular digital bitcell array that is readily reconfigured to a 1-16 bit weight-stationary bit-serial CIM macro. The macro computes parallel dot-product operations between the weights stored in memory and inputs that are serialized from LSB to MSB. Finally, the bit-serial computing scheme significantly reduces the area overhead while sacrificing latency due to bit-by-bit operation cycles. Based on the benefits of digital CIM, reconfigurability, and bit-serial computing architecture, the Colonnade can achieve both high performance and energy efficiency (i.e., both benefits of prior analog and digital accelerators) for processing neural networks. A test-chip with 128 ×128 SRAM-based bitcells for digital bit-serial computing is implemented using 65-nm technology and tested with 1-16 bit weight/input precisions. The measured energy efficiency is 117.3 TOPS/W at 1 bit and 2.06 TOPS/W at 16 bit.

Precise measurement of digital circuit degradation is a key aspect of aging tolerant digital circuit design. In this study, we present a fully digital on-chip reliability monitor for high-resolution frequency degradation measurements of digital circuits. The proposed technique measures the beat frequency of two ring oscillators, one stressed and the other unstressed, to achieve 50 X higher delay sensing resolution than that of prior techniques. The differential frequency measurement technique also eliminates the effect of common-mode environmental variation such as temperature drifts between each sampling points. A 265 X 132 mum²test chip implementing this design has been fabricated in a 1.2 V, 130 nm CMOS technology. The measured resolution of the proposed monitoring circuit was 0.02%, as the ring oscillator in this design has a period of 4 ns; this translates to a temporal resolution of 0.8 ps. The 2 mus measurement time was sufficiently short to suppress the unwanted recovery effect from concealing the actual circuit degradation.

A 2 muW, 100 kHz, 480 kb subthreshold SRAM operating at 0.2 V is demonstrated in a 130 nm CMOS process. A 10-T SRAM cell allows 1 k cells per bitline by eliminating the data-dependent bitline leakage. A virtual ground replica scheme is proposed for logic "0" level tracking and optimal sensing margin in read buffers. Utilizing the strong reverse short channel effect in the subthreshold region improves cell writability and row decoder performance due to the increased current drivability at a longer channel length. The sizing method leads to an equivalent write wordline voltage boost of 70 mV and a delay improvement of 28% in the row decoder compared to the conventional sizing scheme at 0.2 V. A bitline writeback scheme was used to eliminate the pseudo-write problem in unselected columns.

In this work, we present a novel 8T static random access memory (SRAM)-based compute-in-memory (CIM) macro for processing neural networks with high energy efficiency. The proposed 8T bitcell is free from disturb issues thanks to the decoupled read channels by adding two extra transistors to the standard 6T bitcell. A 128 $\times $ 128 8T SRAM array offers massively parallel binary multiply and accumulate (MAC) operations with 64 $\times $ binary inputs (0/1) and 64 $\times $ 128 binary weights (+1/–1). After parallel MAC operations, 128 column-based neurons generate 128 $\times $ 1–5 bit outputs in parallel. The proposed column-based neuron comprises 64 $\times $ bitcells for dot-product, 32 $\times $ bitcells for analog-to-digital converter (ADC), and 32 $\times $ bitcells for offset calibration. The column ADC with 32 $\times $ replica SRAM bitcells converts the analog MAC results (i.e., a differential read bitline (RBL/RBLb) voltage) to the 1–5 bit output code by sweeping their reference levels in 1–31 cycles (i.e., $2^{N}$ –1 cycles for $N$ -bit ADC). The measured linearity results [differential nonlinearity (DNL) and integral nonlinearity (INL)] are +0.314/–0.256 least significant bit (LSB) and + 0.27/–0.116 LSB, respectively, after offset calibration. The simulated image classification results are 96.37% for Mixed National Institute of Standards and Technology database (MNIST) using a multi-layer perceptron (MLP) with two hidden layers, 87.1%/82.66% for CIFAR-10 using VGG-like/ResNet-18 convolutional neural networks (CNNs), demonstrating slight accuracy degradations (0.67%–1.34%) compared with the software baseline. A test chip with a 16K 8T SRAM bitcell array is fabricated using a 65-nm process. The measured energy efficiency is 490–15.8 TOPS/W for 1–5 bit ADC resolution using 0.45-/0.8-V core supply.

Artificial intelligence (AI) and machine learning (ML) are revolutionizing many fields of study, such as visual recognition, natural language processing, autonomous vehicles, and prediction. Traditional von-Neumann computing architecture with separated processing elements and memory devices have been improving their computing performances rapidly with the scaling of process technology. However, in the era of AI and ML, data transfer between memory devices and processing elements becomes the bottleneck of the system. To address this data movement issue, memory-centric computing takes an approach of merging the memory devices with processing elements so that computations can be done in the same location without moving any data. Processing-In-Memory (PIM) has attracted research community’s attention because it can improve the energy efficiency of memory-centric computing systems substantially by minimizing the data movement. Even though the benefits of PIM are well accepted, its limitations and challenges have not been investigated thoroughly. This paper presents a comprehensive investigation of state-of-the-art PIM research works based on various memory device types, such as static-random-access-memory (SRAM), dynamic-random-access-memory (DRAM), and resistive memory (ReRAM). We will present the overview of PIM designs in each memory type, covering from bit cells, circuits, and architecture. Then, a new software stack standard and its challenges for incorporating PIM with the conventional computing architecture will be discussed. Finally, we will discuss various future research directions in PIM for further reducing the data conversion overhead, improving test accuracy, and minimizing intra-memory data movement.