
For the emerging semiconductor transistors, accurate modeling and simulation of these complex devices remain challenging due to their unique physical mechanisms and nanoscale dimensions. This paper presents an easy-use, multi-scale device modeling framework designed to address the challenges of modeling and circuit-level analysis for advanced nanoscale transistors. By integrating physics-based device modeling, automated parameter extraction, and netlist-based circuit simulation, the framework enables simulation from material properties to system-level performance for nanoscale transistors. A key innovation lies in the integration of the Trust Region Reflective (TRR) algorithm with open-source tools to realize a universal, data-training-free parameter extraction workflow, which fills the gap of user-friendly parameter extraction tools for emerging devices. The framework's reliability and flexibility are validated through three representative device cases: FinFETs, graphene transistors, and molybdenum disulfide $\boldsymbol{(} \mathbf{M o S}_{\mathbf{2}} \boldsymbol{)}$ transistors, achieving Root Mean Square Errors (RMSE) of 3.17%, 4.06%, and 4.97% respectively. Circuit-level verification via 19-stage ring oscillators (8.91 GHz operation) and 8T-SRAM modules further confirms its practicality. This framework facilitates rapid prototyping and optimization of emerging transistors, potentially accelerating the transition from lab-scale devices to commercial integrated circuits.
As a high-speed search engine, ternary contentaddressable memory (TCAM) performs fully parallel search operation which activates all cells within each single cycle. However, this results in huge search power consumption especially in voltage-division-based TCAM. To address this issue, we propose a 2-bit encoding scheme and a dynamic power gating architecture for voltage-division-based TCAM. Hybrid simulation is obtained by utilizing a spin transfer torque based MTJ (STT-MTJ) compact model and a 28 nm CMOS design kit. The simulation results demonstrate that the proposed low-power TCAM reaches at least 48.5% power reduction in the worst case compared with conventional designs.
In this paper, we proposed and analyzed a novel four-layer slot integrated optic biosensor for the measurement of temperature. The four-layer configuration is based on siliconon-insulator technology and consists of an intermediate buffer layer. In a four-layer configuration, a thin silicon nitride buffer layer is inserted between the rail and the substrate to enhance the sensitivity, and the cover layer consisting of a bio-analyte to be measured, silicon, and silicon dioxide. In the case of three layer waveguide, the mode confined in the slot influences the sensitivity. Introduction of silicon nitride layer between the oxide and rail layer reduces the refractive index contrast between the substrate and the rail layer resulting in increased penetration depth we obtain a sensitivity $\mathbf{3. 2} \times \mathbf{1 0}^{-4} /{ }^{\circ} \mathbf{C}$.
With the development of the internet of things (IoT), the traditional von-Neumann architecture faces limitations from memory wall, preventing further chip performance improvements. To solve this problem, computing-in-memory (CIM) has been proposed as an effective solution. However, the static storage of weight data in CIM based on non-volatile technology exposes them to malicious attacks. This paper proposes a magnetoresistive random-access memory (MRAM)based multiply-encryption physical unclonable function (PUF) -CIM (Multi-PUFim) to enhance security by encrypting CIM using PUF for multi-encryption obfuscation. The proposed architecture supports both PUF and CIM modes on the same hardware platform. In PUF mode, the design shows strong resistance to modeling attacks, with a prediction accuracy remaining around 50%. In CIM mode, PUF-based encryption reduces the malicious inference accuracy on VGG-8 (CIFAR-10) to approximately $\mathbf{1 0 \%}$. Moreover, the encrypted CIM mode achieves an energy efficiency of approximately 30.498 \~{} 33.489 TOPS/W, effectively balancing security with high computational efficiency.
Emerging planar computing platforms such as atomic-scale computing and silicon photonics combine gates and interconnects in a single 2D plane, making wiring and crossings dominant bottlenecks. This work reframes differentiable logic network training as one-pass synthesis: BitPlanarNet trains a strictly planar gate network that maps one-to-one to device primitives. Building on differentiable logic gate networks (DL-GNs), each neuron selects a 2-input/2-output primitive and connects only to nearest neighbors, yielding a gate-level layout at discretization. It is demonstrated that this concept is physically viable by successfully fabricating a $33 \times 25 ~\text{nm}^{2}$ representative learned dangling-bond logic layout on a hydrogen-passivated silicon surface using a scanning tunneling microscope. On imageclassification tasks, BitPlanarNet achieves accuracy comparable to unconstrained DLGNs-remaining within approximately 1 percentage point on MNIST and 5 percentage points on CIFAR-10-while guaranteeing planar connectivity, providing a versatile methodology for translating learned tasks directly into realizable gate layouts on emerging planar technologies.