隨著神經網路應用的快速擴展，硬體運算需求不斷增加，用記憶體內運算（CIM）來改進的吞吐量和耗能成為了一種有前景的解決方案。這種方法通過在記憶體中直接進行計算，減少了處理器和記憶體之間數據傳輸的時間和耗能。尤其在非揮發性記憶體技術（如電阻式隨機存取記憶體（ReRAM））的發展，即時推論變得更加可行。ReRAM交叉陣列因其能夠在記憶體內進行大量點積操作且具有高效能和低功耗特性，成為深度神經網路推論加速的重要部分。然而，ReRAM的非理想性使得其內存計算容易出錯，這種特性限制了字線並行運算。為了提升性能並增加同時啟動的字線數量，提出了解決方法。

這篇論文提出針對非線性電流累加做電流補償來校準，根據每周期輸入為一的數目去決定補償的多寡，並在ReRAM陣列每周期計算完後進行校準。為了模擬ReRAM陣列所考慮到電路上非理想的電流效應，與在SNN模型上所特有的運算方式，有和業界合作並提取相關電路資料來進行模擬，在準確率97%以上的模擬器上來進行結果的驗證與模擬。

在經由補償校準後的方法，從模擬的結果來看，透過此方法可以增加相比以往將近4倍的加速，在正確率上也能達到98.35%，而且其面積增加不到1.03倍。

As the rapid expansion of neural network applications continues, the demand for hardware computation is increasing. In-memory computing (CIM) emerges as a promising solution to improve throughput and energy efficiency. This approach reduces the time and energy consumption associated with data transfer between processors and memory by performing computations directly within the memory. The development of non-volatile memory technologies, such as resistive random access memory (ReRAM), has made instant inference more feasible. ReRAM crossbar arrays, due to their ability to perform a large number of dot product operations within the memory with high efficiency and low power consumption, have become an essential component in accelerating deep neural network inference. However, the non-ideal characteristics of ReRAM make in-memory computations prone to errors, which limits wordline-level parallelism. To enhance performance and increase the number of concurrently activated wordlines, solutions have been proposed.

In this thesis, we proposes a calibration method for current compensation to address the nonlinear current accumulation. The compensation amount is determined based on the number of inputs equal to one in each cycle, and calibration is performed after each cycle of computation in the ReRAM array. To simulate the non-ideal current effects of the ReRAM array and the specific computation methods in the Spiking Neural Network (SNN) model, industry collaboration was conducted to extract relevant circuit data for simulation. The results were verified and simulated on a simulator with an accuracy of over 97%.

The proposed current compensation and calibration method, based on simulation results, can achieve nearly a fourfold increase in acceleration compared to previous methods, with an accuracy of 98.35%. Additionally, the area increase is less than 1.03 times.

[1] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton, “Imagenet classification with deep
convolutional neural networks”, in Advances in Neural Information Processing Systems, 25,
2012, pp. 1097–1105.
[2] John Backus, “Can programming be liberated from the von neumann style?: A functional
style and its algebra of programs”, in Communications of the ACM, 21(8), 1978, pp. 613–
641.
[3] Abu Sebastian, Manuel Le Gallo, Riduan Khaddam-Aljameh, and Evangelos Eleftheriou,
“Memory devices and applications for in-memory computing”, in Nature Nanotechnology,
15(7). Nature Nanotechnology, 2020, pp. 529–544.
[4] Ali Shafiee, Anirban Nag, Naveen Muralimanohar, Rajeev Balasubramonian, John Paul Strachan,
Miao Hu, R. Stanley Williams, and Vivek Srikumar, “Isaac: A convolutional neural
network accelerator with in-situ analog arithmetic in crossbars”, in 2016 ACM/IEEE 43rd
Annual International Symposium on Computer Architecture (ISCA). IEEE, 2016, pp. 14–26.
[5] Cheng-Xin Xue, Tsung-Yuana Huang, Je-Syu Liu, Ting-Wei Chang, Hui-Yao Kao, Jing-
Hong Wang, Ta-Wei Liu, Shih-Ying Wei, Sheng-Po Huang, Wei-Chen Wei, Yi-Ren Chen,
Tzu-Hsiang Hsu, Yen-Kai Chen, Yun-Chen Lo, Tai-HsingWen, Chung-Chuan Lo, Ren-Shuo
Liu, Chih-Cheng Hsieh, Kea-Tiong Tang, and Meng-Fan Chang, “15.4 a 22nm 2mb reram
compute-in-memory macro with 121-28tops/w for multibit mac computing for tiny ai edge
devices”, in 2020 IEEE International Solid-State Circuits Conference - (ISSCC). IEEE, 2020,
pp. 244—-246.
[6] Ting Chang, Sung-Hyun Jo, and Wei Lu, “Short-term memory to long-term memory transition
in a nanoscale memristor”, in ACS Nano, 5(9). ACS Publications, 2011, pp. 7669–7676.
[7] J. Joshua Yang, JDmitri B. Strukov, and Duncan R. Stewart, “Memristive devices for computing”,
in Nature Nanotechnology, 8(1). Nature Nanotechnology, 2013, pp. 13–24.
[8] Ting-Chang Chang, Kuan-Chang Chang, Tsung-Ming Tsai, Tian-Jian Chu, and Simon M.
Sze, “Resistance random access memory”, in Materials Today Volume 19,Number 5 June
2016, 2016, pp. 254–264.
[9] Miao Hu, John Paul Strachan, Zhiyong Li, Emmanuelle M. Grafals, Noraica Davila, Catherine
Graves, Sity Lam, Ning Ge, Jianhua Joshua Yang, and R. StanleyWilliams, “Dot-product
engine for neuromorphic computing: Programming 1t1m crossbar to accelerate matrix-vector
multiplication”, in 2016 53nd ACM/EDAC/IEEE Design Automation Conference (DAC).
IEEE, 2016.
[10] Wang Kang, Liuyang Zhang, Jacques-Olivier Klein, Youguang Zhang, Dafin´e Ravelosona,
and Weisheng Zhao, “Reconfigurable codesign of stt-ram under process variations in rerambased
neuromorphic systems”, in IEEE Transactions on Nanotechnology, 14(5). IEEE, 2015,
pp. 945–954.
[11] Gang Niu, Hee-Dong Kim, Robin Roelofs, Eduardo Perez, Markus Andreas Schubert, Peter
Zaumseil, Ioan Costina, and ChristianWenger, “Material insights of hfo2-based integrated 1-
transistor-1-resistor resistive random access memory devices processed by batch atomic layer
deposition”, in Scientific reports 6 (2016): 28155. 2016, pp. 1–11, Scientific Reports.
[12] Xiaoyu Sun and Shimeng Yu, “Impact of non-ideal characteristics of resistive synaptic devices
on implementing convolutional neural networks”, in IEEE Journal on Emerging and
Selected Topics in Circuits and Systems(Volume: 9, Issue: 3). IEEE, 2019, pp. 570–579.
[13] Pai-Yu Tan, Po-Yao Chuang, Yen-Ting Lin, Cheng-Wen Wu, and Juin-Ming Lu, “A powerefficient
binary-weight spiking neural network architecture for real-time object classification”,
in arXiv:2003.06310, 2020.
[14] Po-Yao Chuang, Pai-Yu Tan, Cheng-Wen Wu, and Juin-Ming Lu, “A 90nm 103.14 tops/w
binary-weight spiking neural network cmos asic for real-time object classification”, in 2020
57th ACM/IEEE Design Automation Conference (DAC). IEEE, 2020, pp. 1–6.
[15] Kuan-Wei Hou, Hsueh-Hung Cheng, Chi Tung, Cheng-Wen Wu, and Juin-Ming Lu, “Fault
modeling and testing of memristor-based spiking neural networks”, in 2022 IEEE International
Test Conference (ITC)). IEEE, 2022, pp. 92–99.
[16] Rajeev Balasubramonian, Andrew B. Kahng, Naveen Muralimanohar, Ali Shafiee, and Vaishnav
Srinivas, “Cacti 7: New tools for interconnect exploration in innovative off-chip memories”,
in ACM Trans. Archit. Code Optim., vol. 14, no. 2, jun 2017, 2017.