Exploiting Dual-Gate Ambipolar CNFETs for Scalable Machine Learning Classification

Abstract

Ambipolar carbon nanotube based field-effect transistors (AP-CNFETs) exhibit unique electrical characteristics, such as tri-state operation and bi-directionality, enabling systems with complex and reconfigurable computing. In this paper, AP-CNFETs are used to design a mixed-signal machine learning logistic regression classifier. The classifier is designed in SPICE with feature size of 15 nm and operates at 250 MHz. The system is demonstrated in SPICE based on MNIST digit dataset, yielding 90% accuracy and no accuracy degradation as compared with the classification of this dataset in Python. The system also exhibits lower power consumption and smaller physical size as compared with the state-of-the-art CMOS and memristor based mixed-signal classifiers. Power consumption and physical size of integrated circuits (ICs) is an increasing concern in many emerging machine learning (ML) applications, such as, autonomous vehicles, security systems, and Internet of Things (IoT). Existing state-of-the-art architectures for digital classification, are highly accurate and can provide high throughput 1. These classifiers, however, exhibit high power consumption and occupy a relatively large area to accommodate complex ML models. Alternatively, mixed-signal classifiers have been demonstrated to exhibit orders of magnitude reduction in power and area as compared to digital classifiers with prediction accuracy approaching the accuracy of digital classifiers 2-9. While significant advances have been made at the ML circuit and architecture levels (e.g., in-SRAM processing 2 , comparator based computing 3 , and switched-capacitor neurons 4), the lack of robust, ML-specific transistors is a primary concern in ML training and inference with all conventional CMOS technologies. To efficiently increase the density and power efficiency of modern ML ICs while enabling complex computing, emerging technologies should be considered. While the non-volatility of memristors has proven quite attractive for storing the weights required for feature-weight multiplication 10-13 , field-effect transistors provide several advantages over memristors for ML. First, transistors provide a broader range of linear tuning of resistance, thereby better matching ML models. Second, transistors are not subject to the deleterious aging that deteriorates memristor behavior over time. Finally, the connectivity between feature-weight multiplication layers requires electrical signal gain, which cannot be provided by memristors; transistors can be used for such interlayer connections, thereby enabling a monolithic integrated circuit that can be fabricated efficiently. Of particular interest for on-chip classification are ambipolar devices. Owing to unique electrical characteristics, as described in the next section, ambipolar devices are expected to provide efficient on-chip training and inference solutions and reduce design and routing complexity of ML circuits. A carbon nanotube (CNT) ambipolar device has been reported as a potential candidate for controllable ambipol...

The design and performance of different nanoelectronic binary multipliers

“…It is possible to address the first two criteria listed above by using different nanodevices such as the carbon nanotube FET (CNFET) [2], resonant tunnel diode (RTD) [3], rapid single quantum flux (RSQF) devices [4], single-electron transistor (SET) [5][6][7][8], etc. The current work considers different potential nanoelectronic binary multiplier design based on the single-electron threshold logic gate (SE-TLG) [9][10][11][12], the hybrid SET-CMOS approach [13][14][15], and CNFET-based designs [16,17]. The results are also compared with the conventional CMOS-based implementation of the same.…”

Cascaded Logic Gates Based on High-Performance Ambipolar Dual-Gate WSe2 Thin Film Transistors

“…This has been widely demonstrated in organic ambipolar transistors − and 2D material transistors. − Among the large atomically thin material families, graphene, carbon nanotube (CNT), black phosphorus (BP), and certain TMDs like WSe 2 exhibit intrinsic ambipolar behavior without the need for doping. The primary challenge of ambipolar transistors is their higher off-state current compared to unipolar devices, particularly for small-bandgap materials such as graphene and BP. , Consequently, a dual-gate structure is advantageous in ambipolar devices, as it not only suppresses the off-state current but also enables reconfigurable ambipolar TFTs that can be reversibly switched between p-type and n-type modes. − Furthermore, due to the independent input of the two gates, logic circuits made with ambipolar dual-gate TFTs can potentially achieve the desired operation with fewer transistors and lower power consumption than complementary metal–oxide semiconductor (CMOS) technology. − …”

“…CNT-array-based devices demonstrate outstanding ambipolar behavior, but they face the challenge of achieving high current density due to their 1D nature and the purification of semiconductive CNTs. CNT-network-based devices possess better compatibility and simplicity in fabrication, but they suffer from relatively low mobility. , BP devices have shown promise in small supply voltage logic circuits, but they exhibit a relatively high off-state current, even with the dual-gate structure, due to the small bandgap of ∼0.3 eV. The threshold voltage ( V T ) of the demonstrated BP devices shifts from 0 V, leading to increased static power consumption.…”

A Single-MOSFET Analog High Resolution-Targeted (SMART) Multiplier for Machine Learning Classification

“…Recent state-of-the-art mixed-signal classifiers typically exhibit accuracy of 90%-99% and the overall energy consumption in the range of hundreds of picojoules to hundreds of nanojoules per decision for typical image recognition datasets [7]- [13]. Emerging device technologies are also being considered for providing power and area efficient alternatives for the conventional CMOS based classifiers [14]. Accuracy of 90% and energy of 25 pJ per decision has been recently reported in [15], [16].…”