Computing-In-Memory (CIM), based on non-von Neumann architecture, has lately received significant attention due to its lower overhead in delay and higher energy efficiency in convolutional and fully-connected neural network computing. Growing works have given the priority to researching the array of memory and peripheral circuits to achieve multiply-and-accumulate (MAC) operation, but not enough attention has been paid to the high-precision hardware implementation of non-linear layers up to now, which still causes time overhead and power consumption. Sigmoid is a widely used non-linear activation function and most of its studies provided an approximation of the function expression rather than totally matched, inevitably leading to considerable error. To address this issue, we propose a high-precision circuit implementation of the sigmoid, matching the expression exactly for the first time. The simulation results with the SMIC 40 nm process suggest that the proposed circuit implemented high-precision sigmoid perfectly achieves the properties of the ideal sigmoid, showing the maximum error and average error between the proposed simulated sigmoid and ideal sigmoid is 2.74% and 0.21%, respectively. In addition, a multi-layer convolutional neural network based on CIM architecture employing the simulated high-precision sigmoid activation function verifies the similar recognition accuracy on the test database of handwritten digits compared to utilize the ideal sigmoid in software, with online training achieving 97.06% and with offline training achieving 97.74%.

In this article, a low-cost quadruple-node-upsets resilient latch (LCQRL) design is proposed. To meet the high-reliability demands of safety-critical applications, the latch integrates nine soft-error-interceptive modules (SIMs) to form robust feedback loops, ensuring complete resilience to quadruple-node upsets (QNUs). Each Sim comprises ten CMOS transistors and a clocked inverter. Notably, C-element (CE) and dual interlocked storage cell (DICE) modules are not employed in this circuit, resulting in a small area and low power consumption. The simulation results verify the complete QNU self-recoverability and cost-effectiveness of this design. Compared with the existing radiation-hardened QNU resilient latches, the LCQRL latch demonstrates significant improvements in area, power consumption, and area–power–delay product (APDP) by 47.8%, 63%, and 75.5%, respectively. Furthermore, it exhibits low sensitivity to process, voltage, and temperature (PVT) variations.

In this paper, a fast-transient-response NMOS low-dropout regulator (LDO) with a wide load-capacitance range was presented to provide a V/2 read bias for cross-point memory. To utilize the large dropout voltage in the V/2 bias scheme, a fast loop consisting of NMOS and flipped voltage amplifier (FVA) topology was adopted with a fast transient response. This design is suitable to provide a V/2 read bias with 3.3 V input voltage and 1.65 V output voltage for different cross-point memories. The FVA-based LDO designed in the 110 nm CMOS process remained stable under a wide range of load capacitances from 0 to 10 nF and equivalent series resistance (ESR) conditions. At the capacitor-less condition, it exhibited a unity-gain bandwidth (UGB) of approximately 400 MHz at full load. For load current changes from 0 to 10 mA within an edge time of 10 ps, the simulated undershoot and settling time were only 144 mV and 50 ns, respectively. The regulator consumed 70 µA quiescent current and achieved a remarkable figure-of-merit (FOM) of 1.01 mV. At the ESR condition of a 1 µF off-chip capacitor, the simulated quiescent current, on-chip capacitor consumption, and current efficiency at full load were 8.5 µA, 2 pF, and 99.992%, respectively. The undershoot voltage was 20 mV with 800 ns settling time for a load step from 0 to 100 mA within the 10 ps edge time.

Emerging applications such as edge AI, the Internet of Things (IoT), and mobile computing demand memory technologies that deliver both ultralow power consumption and high throughput. Phase-change memory (PCM) is a promising candidate due to its scalability and CMOS compatibility, but the conventional 1-transistor–1-resistor (1T1R) architecture suffers from narrow read margins and high programming energy. In this work, a differential 3-transistor–2-resistor (3T2R) PCM cell is evaluated, which enhances read robustness through differential comparison while maintaining a compact layout. Compared with 1T1R, the 3T2R prototype achieves an 86.7% reduction in write energy and a $5.2\times $ improvement in write speed. Endurance and retention exhibit reliable performance under low-to-moderate cycling, while the cell area offers a balanced compromise between compactness and differential robustness. These results establish the 3T2R architecture as an efficient solution for embedded and edge memory applications where energy efficiency and high-speed operation are critical.