搜索

x
中国物理学会期刊

基于忆阻器交叉阵列的低开销存内加法器及加解密电路设计

CSTR:32037.14.aps.75.20260679

Design of a low-overhead in-memory adder and encryption-decryption circuit based on memristor crossbar array

CSTR:32037.14.aps.75.20260679
PDF
导出引用
  • 针对传统冯·诺依曼架构存算分离导致的存储墙瓶颈, 以及现有忆阻器全加器方案存在器件开销大、运算步骤多、硬件复杂度高等问题, 本文基于忆阻器交叉阵列开展新型存内加法器设计. 首先, 提出一种低开销、少步骤的1位全加器, 仅需6个忆阻器和2步操作即可完成求和与进位输出, 在器件数量与运算效率上优于现有方案. 其次, 基于该全加器构建存内加解密电路, 实现存储、运算与加解密功能的一体化融合, 可满足安全敏感应用场景的需求. 进一步, 通过全加器设计n位行波进位加法器, 经品质因数(figure of merit, FoM)评估, 在\mathrmFoM_\mathrmB(步骤数和器件数同等权重)和\mathrmFoM_\mathrmS(侧重步骤数)指标下性能改进分别达到1.37\times — 12.18\times和1.80\times— 41.98\times. 本文设计为基于忆阻器的存算一体加法电路的高效实现提供了可行方案.

    The adder is a core arithmetic component in modern microprocessors, cryptographic circuits, and other systems. Most existing memristor-based full adders adopt the cascaded discrete basic logic gate approach, which leads to large device overhead and long operation steps, making it difficult to meet the requirements of high-speed data processing and real-time encryption. To address these issues, this paper first designs a low-overhead full adder circuit based on the memristor array, which consists of input memristors, output memristors, and an auxiliary memristor. The entire computation is performed within the memristors, using the high-resistance state and low-resistance state to represent logic 0 and logic 1, respectively. Based on Ohm’s law, Kirchhoff’s law, and the full adder truth table, a detailed derivation of the circuit is carried out, and the operating voltages are determined. During computation, specific operating voltages are applied to the bit lines of involved memristors, triggering conditional resistance switching of the output memristor determined by inputs. The proposed full adder requires only six memristors and two operation steps to produce the sum and carry, outperforming existing schemes in both device count and operation efficiency.
    Based on the above one-bit full adder, this paper further adopts a one-transistor-one-memristor (1T1M) array to construct an in-memory encryption-decryption circuit. The circuit uses a single-bit full adder as the basic unit to form an encryption-decryption unit (EDU), and an m\times n array of EDUs is integrated to form the encryption-decryption array. For image encryption and decryption applications, two operating modes are developed: 1) Semi-parallel mode: a single column of EDUs is activated at a time to process all pixels in one column of the image in parallel, completing encryption or decryption column by column. Processing an m\times n binary image requires only 2n operation steps. 2) Fully parallel mode: by disconnecting the columns of the array through switches, all columns of EDUs can simultaneously process all columns of pixels, completing the encryption or decryption of the entire image in only 2 steps. Verification using 256\times 256 binary images shows, by comparing the images before and after encryption and decryption, that the proposed encryption-decryption circuit can correctly perform image encryption and decryption.
    Finally, this paper proposes a COPY logic operation based on the 1T1M crossbar array, which can copy the resistance state of one memristor to another. By integrating the proposed full adder and COPY gate, an n-bit ripple-carry adder is designed. Specifically, the output carry c_i of the (i-1)-th bit is calculated serially and copied to the memristor at the i-th row of the same column via COPY operation. This serial computation and copying process is iterated until all carries c_i+1 (i = 1, 2, 3,\cdots) are generated. Subsequently, the odd-indexed sums s_i (i = 1, 3, 5,\cdots) and the even-indexed sums s_i (i = 2, 4, 6,\cdots) are computed in parallel, respectively. The complete operation of this n-bit ripple-carry adder requires only 2n+1 operation steps. Compared with existing works, the proposed design achieves improvements of 1.37× to 12.18× in \rm FoM_\mathrmB (equally weighted by step and device count) and 1.80× to 41.98× in \rm FoM_\mathrmS (weighted toward step count), exhibiting distinctly superior overall performance.

    目录

    返回文章
    返回
    Baidu
    map