@inproceedings{88c8fe108fd64ccd93f8d5f94eb4937f,
title = "In-Memory Computing Primitive for Sensor Data Fusion in 28 nm HKMG FeFET Technology",
abstract = "In this work, we exploit the spatio-temporal switching dynamics of ferroelectric polarization to realize an energy-efficient, and massively-parallel in-memory computational primitive for at-node sensor data fusion and analytics based on an industrial 28nm HKMG FeFET technology [1]. We demonstrate: (i) the spatio-temporal dynamics of polarization switching in HfO2-based ferroelectrics under the stimuli of sub-coercive voltage pulses using experiments and phase-field modeling; (ii) an inherent rectifying conductance accumulation characteristic in FeFET with a large dynamic range of G/G > 100 in the case of 3.0V, 50ns gate pulses; (iii) transition to more abrupt accumulation characteristics due to single/few domain polarization switching in scaled FeFET (34nm LG); and (iv) successful detection of physiological anomalies from realworld multi-modal sensor data streams.",
author = "K. Ni and B. Grisafe and W. Chakraborty and Saha, \{A. K.\} and S. Dutta and M. Jerry and Smith, \{J. A.\} and S. Gupta and S. Datta",
note = "Funding Information: The physiological signals of patient 221 from the MIMIC database show three anomalies (Fig.13(a)) [7]. By translating the real signals to gate pulses to FeFET, a conductance read can differentiate the anomalies from normal signals reliably as the anomalies are highly correlated among different signals. A majority voting on the number of FeFET conductance over the threshold identifies the anomaly faithfully. Fig.13(b) shows that detection based on 3 signals cause information loss while in the case of 7 signals the detection is successful as more global information is extracted. An array level implementation with the FeFET cells is shown in Fig. 14. Compared with a CMOS ASIC based counter [8] or PCM cell [2], FeFET exhibits high speed, large Gmax/Gmin ratio and low write energy (Fig.15). CONCLUSIONS In summary, we demonstrate an in-memory computational primitive for statistical correlation detection based on an industrial 28nm HKMG FeFET technology. The spatiotemporal polarization switching dynamics, involving the domain nucleation and growth, are responsible for polarization accumulation, which causes continuous conductance accumulation in large FeFET. Extremely scaled FeFET exhibits abrupt switching but can be compensated by multiple devices grouping. Fast operation with 3.0V, 50 ns is also demonstrated. The conductance accumulation characteristic is successfully applied to detect real-world physiological signal anomalies. These results make FeFET based correlation detection an ultradense, highly energy-efficient, and massively parallel system for real-time signal processing for sensor analytics. ACKNOWLEDGEMENT We would like to thank M. Trentzsch, S. Dunkel, S. Beyer, and W. Taylor at Globalfoundries Dresden, Germany for providing 28nm HKMG FeFET test devices. This work was supported in part by the Semiconductor Research Corporation (SRC) and DARPA. REFERENCES [1] M. Trentzsch et al., IEDM 2016 [2] A. Sebastian, Nature Comm. 2017 [3] H. Mulaosmanovic et al., Appl. Mater. Interfaces 2017 [4] Y. H. Shin, et al., Nature 2007 [5] H. Mulaosmanovic et al., Appl. Mater. Interfaces 2018 [6] B. R. Gaines et al., 1969 [7] G. B. Moody et al., Computers in Cardiology 1996 [8] S. Mathew, ISSCC, 2004 Publisher Copyright: {\textcopyright} 2018 IEEE.; 64th Annual IEEE International Electron Devices Meeting, IEDM 2018 ; Conference date: 01-12-2018 Through 05-12-2018",
year = "2018",
month = jul,
day = "2",
doi = "10.1109/IEDM.2018.8614527",
language = "English (US)",
series = "Technical Digest - International Electron Devices Meeting, IEDM",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
pages = "16.1.1--16.1.4",
booktitle = "2018 IEEE International Electron Devices Meeting, IEDM 2018",
address = "United States",
}