A recent surge of research on deep convolutional neural network (DCNN) has given a challenge on existing computation architecture which has the flows in speed and memory bottleneck. In the pretreatment of deep learning, mask operation is frequently used to remove noise or fetch information. However, under data-intensive conditions, applying mask frequently can put an extremely heavy memory/communication burden on computing system. An efficient substrate of DCNN for ameliorating mask operation is urgently needed. In this paper, we present a self-renewing mask circuit (SRMC) to alleviate/solve earlier problems. First, a new approach to apply mask is presented based on computation-in-memory architecture that implements high-performance processor and high-density memory in the same physical location. Second, we designed the peripheral circuit which can provide self-renewing function to further avoid the data exchange with an external space. As opposed to most other computational element, which calculates two 1-bit data, the proposed SRMC storage multibit value and calculate several of them parallel. The calculating ability of SRMC is significantly superior to those of computational element in general architectures. Moreover, mean filter and edge detector are implemented to illustrate the effectiveness and scalability of the proposed circuit. Finally, we discussed two possible methods to enhance the practicability of our scheme.
15 May 2020 in IEEE Journal on Exploratory Solid-State Computational Devices and Circuits
Linear programming optimization is central to engineering designs, logistics management, and decision making in every sector of the economy. Traditional hardware using CPU and GPU platforms for this purpose is severely limited by the scaling of the transistor technology. In this paper, we design an analog in-memory computation circuit for accelerating linear programming optimization problems. The scheme includes a memristive crossbar array and analog peripheral circuits that do not require DAC/ADC between each algorithm iteration. In addition, several key parameters related to non-ideal device characteristics and interconnect parasitics are discussed for providing practical guidelines. Furthermore, three design schemes are proposed to alleviate the computation error caused by the interconnect resistance for a large-scale crossbar array implementation. Optimal design parameters are quantified under a given number of array size and memristive resistance. Finally, the proposed hardware accelerator and error mitigation techniques are applied to six real-world power system optimization problems. The results show that the average error of generator power and the overall cost is less than 3%. It is demonstrated that the proposed accelerator achieves area, delay and, energy consumption reductions of ~151×, ~ 33×, and ~ 21×, respectively, compared to the CMOS digital circuits at the 16nm technology node for a 1000×1000 array with a 6-bit precision.
The use of multi-output look-up tables (LUTs) is a widely adopted approach in contemporary commercial field-programmable gate arrays (FPGAs). Larger LUT configurations (e.g., six-input LUTs) can be partitioned into smaller LUTs (e.g., two five-input LUTs, maintaining a total input count of less than six). This capability of generating a second output from a larger LUT is not only crucial for reducing logic cell count and enhancing the utilization efficiency of logic resources—thus conserving area—but also plays a key role in optimizing system-level delays and energy consumption. In this paper, we propose an efficient multi-output LUT mapping technique, incorporating several highly efficient technology mapping algorithms, which focus on optimizing the mapping from an interconnection perspective as alternatives to directly merging smaller LUTs. These algorithms include a side-fanout insertion algorithm, and a runtime multi-output cut generation algorithm. The proposed methods improve mapping efficiency and enhance performance. The benchmarking results demonstrate that the dual-output mapping algorithms achieve LUT area reductions of up to 35% and 6%, compared to the state-of-the-art ABC six-input, single-output LUT mapping technique and previous work focusing on dual-output LUT mapping techniques that optimize cut generation parameters. Moreover, FPGA system-level simulations also show that area, delay, and energy can all be optimized based on this multi-output mapping technique.
6 March 2023 in IEEE Open Journal of the Computer Society
In this article, we propose a generic design methodology to achieve area-efficient reconfigurable logic circuits by using exact synthesis based on Boolean satisfiability (SAT) solver. The proposed methodology better leverages the high representation ability of emerging reconfigurable logic gates (RLGs) to achieve reconfigurable circuits with fewer gates. In addition, we propose a fence-based acceleration method to provide >10× speed up for the synthesis without an observable loss of optimality. Furthermore, four sets of RLGs are developed based on a recently proposed valley-spin device as a case study to demonstrate the advantage of the proposed circuit. Simulations have been performed to analyze the impact of the fence searching algorithm and combination of operators. Based on disjoint-support decomposable (DSD) benchmarks, up to 38% and 73% reductions are observed in the area and energy-delay-area product (EDAP), respectively, compared to CMOS counterparts. Compared to the two existing synthesis methods, the proposed scheme provides 40% and 26.3% reduction in EDAP based on MCNC benchmark.
Datapath synthesis is a crucial step in synthesis flow and aims at globally minimizing an area by identifying shareable logic structures. This paper introduces a novel Directed Acyclic Graph (DAG)-based datapath synthesis method based on graph isomorphism and gate reconfiguration. Unlike algorithms that identify common specification logic, our approach simplifies the problem by focusing on searching for common topology. Leveraging the concept of gate reconfiguration, our algorithm extends the applicability of DAG-based datapath synthesis by transforming a topology-equivalent network into a specification-equivalent network. Experimental results demonstrate up to 23.6% improvement when optimizing the adder–subtractor circuit, a scenario not addressed by existing DAG-based datapath synthesis algorithms.
16 June 2025 in IEEE Transactions on Emerging Topics in Computing
Reconfigurable devices are attracting growing interest as both a potential alternative and complement to traditional CMOS technology. This paper develops a novel fieldprogrammable gate array (FPGA) architecture based on MClusters, which is made of fast and area-efficient 2-input lookup tables (LUTs) through reconfigurable field-effect transistors (RFETs). To fully utilize the MClusters, we propose an SAT-based delay-aware packing algorithm for the technology mapping. In addition, we integrate a partitioning algorithm to divide the circuit into several sub-circuits to further reduce the global routing resources and their associated switching energy of the system. Finally, we develop an efficient technology/circuit/system co-design framework for optimizing the overall performance of FPGAs. Based on comprehensive benchmarking, results demonstrate that optimal design yields significant reductions of up to 39% area, 36% wire length, and 40% switching energy compared to traditional CMOS 6-input LUT FPGAs.