Md Samzid Bin Hafiz1,Md Tashfiq Bin Kashem1,ABM Hasan Talukder1,Raihan Sayeed Khan1,Helena Silva1,Ali Gokirmak1
University of Connecticut1
Md Samzid Bin Hafiz1,Md Tashfiq Bin Kashem1,ABM Hasan Talukder1,Raihan Sayeed Khan1,Helena Silva1,Ali Gokirmak1
University of Connecticut1
3D integration of electronic memories led to enormous storage capacity with substantially reduced cost, new processor architectures, and artificial neural network and artificial intelligence (AI) implementations dramatically improved performance and application space. However, the speed of typical data-heavy computations are still limited by memory access latencies between CPU and DRAM (dynamic random access memory), flash memory and magnetic drive<sup>[1]</sup>. The next stage of improvements is expected to come from monolithic integration of memory and logic, and complementing silicon CMOS with new capabilities, which is likely to include devices with dynamic materials. The emerging resistive non-volatile memory (RRAM) technologies, such as phase change memory (PCM)<sup>[2]</sup>, conductive bridging RAM (CBRAM)<sup>[3]</sup> and magnetic RAM (MRAM)<sup>[4], [5]</sup>, offer the possibility of monolithic integration of processor and high-performance non-volatile memory.<br/><br/>PCM devices utilize reversible changes in resistance of a small volume of material (1000s of atoms) as it is thermally switched between amorphous and crystalline phases, which makes PCM (i) operable using signal polarity for set and reset, (ii) virtually immune to read disturbance, (iii) insensitive to point defects and thermodynamically stable down to ~ (5 nm)<sup>3</sup> scale, (iv) have a large memory window (orders of magnitude) at a very desired resistance range for CMOS<sup>[6]</sup>.<br/><br/>The resistance of the reset (amorphous) state of the PCM cells increases over time, widening the memory window but hinders multi-bit-per-cell PCM. Although resistance drift has been commonly attributed to structural relaxation, resistance drift at lower temperatures (below 100 K) and its response to high-field stress and photoexcitation suggest an electronic origin to resistance drift. This is a promising finding for the potential mitigation strategies for resistance drift and the viability of multi-bit implementations of PCM<sup>[7], [8]</sup> including neuromorphic computing.<br/><br/>In this study, we use our experimental results to construct a finite element model that captures electronic transport in amorphous Ge<sub>2</sub>Sb<sub>2</sub>Te<sub>5</sub> (GST)<sup>[9]</sup> and investigate potential mitigation strategies. We model local hole-traps that are activated in time with a temperature dependent probability. The remaining fixed charges lead to local potential variations, which then give rise to formation of percolation paths in time. The changes give rise to resistance fluctuations (noise) along with the increase in resistance. Once the steady state is reached, resistance drift stops. Our experimental results show that the resistance drift can be dramatically accelerated with application of high-field stresses, which can be explained by acceleration of activation and removal of trapped holes in the amorphous material.<br/><br/>Acknowledgment: This work is supported by the US National Science Foundation award ECCS 1711626.<br/>References<br/>[1] M. K. Qureshi, V. Srinivasan, and J. A. Rivers, <i>ACM SIGARCH Comput. Archit. News</i>, vol. 37, no. 3, 24–33, <b>2009</b> .<br/>[2] S. R. Ovshinsky, <i>Phys. Rev. Lett.</i>, vol. 21, no. 20, 1450–1453, <b>1968</b> .<br/>[3] M. N. Kozicki, M. Park, and M. Mitkova, <i>Nanotechnology, IEEE Trans.</i>, vol. 4, no. 3, 331–338, <b>2005</b> .<br/>[4] B. N. Engel <i>et al.</i>, <i>Magn. IEEE Trans.</i>, vol. 41, no. 1, 132–136, <b>2005</b> .<br/>[5] A. Deschenes, S. Muneer, M. Akbulut, A. Gokirmak, and H. Silva, <i>Beilstein J. Nanotechnol.</i>, vol. 7, no. 1, <b>2016</b> .<br/>[6] P. Cappelletti, R. Annunziata, F. Arnaud, F. Disegni, A. Maurelli, and P. Zuliani, <i>J. Phys. D. Appl. Phys.</i>, vol. 53, no. 19, 193002, <b>2020</b> .<br/>[7] R. S. Khan, F. Dirisaglik, A. Gokirmak, and H. Silva, <i>Appl. Phys. Lett.</i>, vol. 116, no. 25, 253501, <b>2020</b> .<br/>[8] R. S. Khan, A. H. Talukder, F. Dirisaglik, H. Silva, and A. Gokirmak, <i>arXiv Prepr. arXiv2002.12487</i>, <b>2020</b> .<br/>[9] A. Pirovano, A. L. Lacaita, A. Benvenuti, F. Pellizzer, and R. Bez, <i>IEEE Trans. Electron Devices</i>, vol. 51, no. 3, 452–459, <b>2004</b> .