With the emerging transition from programmed to learning computing, traditional electronic ICs are unlikely to be able to match the massive perceptual data that will need to be processed in real-time. Neuromorphic engineering may surmount the von Neumann bottleneck by creating high-performance hardware for distributed and parallel processing with low power consumption. Nowadays, most of high-efficiency neuromorphic hardware is based on artificial synapsis and neuronal devices, utilizingelectrical or optical platforms and thus correspondingly limited in terms of bandwidth or footprint, respectively. In this work, we introduce a plasmon-stimulated synaptic transistor concept that may combine the high bandwidth of a photonic network with the compactness of electronic circuits. The actual synaptic characteristic, including excitatory postsynaptic current and paired-pulse facilitation, is achieved utilizing charge trapping effects to control the conductivity of a monolithic Al-Ge-Al Schottky barrier field-effect transistor. The population of traps is thereby controlled by hot electrons arising from the non-radiative decay of deliberately applied surface plasmon polariton spikes. The temporal dynamics of the trapping process and thus the plasticity of the synaptic transistor can be set by the geometry and operation mode of the synaptic transistor as well as the intensity and duration of the plasmon spikes.
Article
Plasmon-Stimulated Synaptic Transistor
https://doi.org/10.21203/rs.3.rs-2742680/v1
This work is licensed under a CC BY 4.0 License
Version 1
posted
You are reading this latest preprint version
synaptic transistor
plasmon
Schottky barrier field effect transistor
germanium
aluminum
Today’s computer hardware is based on the von Neumann architecture, with the processor and memories as separate components connected - or separated - by the system bus. The transfer of data inevitably consumes energy and increases computing times limited by the speed of the system bus.1,2 This effect was counteracted for a long time by the scaling approach, to continue the Power-Performance-Area-and-Cost scaling3 which has replaced the traditional More-Moore scaling. Thereby processor speeds became significantly faster,4 and memory supported greater densities,5 but transfer rates between the processor and memory have made only modest gains.6,7 At the same time, emerging distributed computing paradigms such as the Internet of Things place even higher demands on hardware performance,8 forcing research towards the exploration of new computer architectures.9,10 Aside of quantum computing,11 neuromorphic circuits are considered as an alternative to surmount the von Neumann bottleneck.12,13 Neuromorphic computing models the way the brain works, with the part of synapses played by either software or an artificial device that exhibits excitatory and inhibitory behavior. A synaptic transistor device can take several states rather than 1 or 0, thus effectively mimicking the way the strength of a connection between two synapses can vary.14–16 Changing those weights in artificial synapses enables the brain-inspired neuromorphic systems to learn.17,18
Currently, most synaptic devices are electrically stimulated19–22 finally afflicted with the problem of narrow bandwidth and high energy consumption. In this regard, neuromorphic photonics has come forward, which aims to transfer the high-bandwidth and low-energy interconnect benefits of photonic circuitry to neuromorphic platforms.23–27 However, obvious obstacles include the diffraction limited integration and miniaturization capability, far poorer than those achievable in modern electronics as well as a weak light-matter interaction typical for existing devices. Light-stimulated synaptic transistors would have to meet footprint and efficiency requirements similar to their electronic counterparts calling for extreme capabilities to generate, guide, modulate and detect photons.
Plasmonics, has emerged as one potential solution for high-performance integrated on-chip circuits that can combine high speeds, the bandwidth of a photonic network and ultra-compact circuit architectures.28–31 Surface plasmon polaritons (SPPs) are electromagnetic waves propagating along a metal-dielectric interface and are coupled to collective oscillations of free electrons in the metal. As the optical modes of SPPs can be significantly smaller than the wavelength of light, plasmonic devices can even be interfaced with similarly sized nanoscale electronics. Moreover, metallic SPP waveguides may even perform a dual function and simultaneously carry both optical and electrical signals,32 giving rise to exciting new capabilities.33 With regard to scaling and integration, notably with the maturity of electromagnetic simulations and current complementary metal-oxide semiconductor (CMOS) fabrication techniques, a variety of such plasmonic designs can be fabricated in a Si foundry right now.34–36
In this work, we demonstrate a plasmon-stimulated synaptic device utilizing charge trapping effects to modulate the electrical characteristic of a synaptic Al-Ge-Al Schottky barrier field-effect transistor (SBFET). Aluminum (Al) is excellent for plasmonics because of a broad response spectrum from ultraviolet to infrared, a self-limited native oxide layer protecting the metal surface and CMOS-compatibility.37 Current Si-based ICs technology already uses nanoscale Al interconnects, to route electronic signals between transistors on a chip.38 Germanium (Ge) as the channel material of the synaptic SBFET is advantaged by its full compatibility with CMOS technology to eventually create a monolithic solution. The synaptic characteristic of the actual device such as excitatory post synaptic current (EPSC) or paired pulse facilitation (PPF) is determined by the properties of Ge or more precisely the intrinsic traps at the Ge/GeOx interface and in the oxide layer. These traps capture hot electrons induced by the non-radiative decay of SPP spikes which then has a feedback effect on the electrical conductivity of the SBFET, responding in a neuron-/synapse‐like fashion.
2.1 Device architecture and operation mode of plasmon-stimulated transistor
The layout of the proposed plasmon-stimulated synaptic transistor is schematically depicted in Fig. 1a with the respective top-view scanning electron microscopy (SEM) image in the inset. The device essentially comprises a monolithic Al-Ge-Al heterostructure39 on top of an oxidized Si wafer resembling a back-gated SBFET. The overall fabrication sequence is described in detail in the supporting information (Figure S1). The Ge channel is connected with monolithic Al feed lines to macroscopic pads of source (S) and drain (D). The buried oxide (150 nm SiO2) and p-doped Si handle wafer of the Germanium-on-insulator substrate represent the gate dielectric and common back-gate (G), respectively. Two oval recesses in the Al feed lines act as simplified plasmon couplers for launching SPPs when irradiated with a laser beam linearly polarized perpendicular to the oval cutouts.
The overall working principle of the plasmon-stimulated synaptic transistor is explained by Fig. 1a:
① When the plasmon coupler is irradiated with a laser beam, SPPs are launched and propagate along the Al metallization towards the Al-Ge junction. The Al feed line thus not only acts as an electrical contact for the SBFET, but also as a metal strip SPP waveguide.40
② SPPs approaching the Al-Ge interface partially decay and hot electrons with sufficient energy are injected above the effective Schottky barrier ΦB into the Ge channel of the SBFET. Notably, the hot carrier’s momentum must be in the direction to the Al-Ge junction, so that the kinetic energy component in that direction is sufficient to surpass the Schottky barrier.41 Otherwise, they will be reflected back and dispersed by scattering between free electrons or phonons in Al. The difficulty of achieving both, energy and momentum matching is in general a main cause of the low quantum efficiency for extracting excited carriers in plasmonic devices.42 The architecture of the actual device is thus ideal as the SPPs propagate in plane of the monolithic heterostructure towards the Al-Ge junction.43
③ Hot electrons surpassing the effective Schottky barrier and entering the Ge segment are subject to an applied bias (VD) resulting in a measurable current further denoted as plasmon current (IPl). Some of the injected high energetic electrons occupy traps, which then negatively charged have a feedback effect on the electrical conductivity of the SBFET. The dynamics of occupation and depletion of fast traps at the Ge/GeOx interface or slow traps in the oxide layer and the respective feedback on the conductivity of the SBFET ultimately induces the synaptic behavior.
2.2 Electro-optical characterization.
Before proving synaptic functionality we demonstrate how SPPs induced modifications of the trap population modulates the transconductance of the SBFET. Figure 1b illustrates transfer characteristics of the SBFET shown in the upper inset of Fig. 1a, measured in the dark and with the plasmon coupler irradiated with a λ = 785 nm laser beam. The strong increase of the drain current ID with negative gate voltage observed in the dark is typical for thin film Ge enhancement mode SBFETs due to surface doping.44 For the actual 75 nm thick channel, imperfections at the Ge/GeOx interface and negatively charged traps in the oxide determine the p-type response with holes as majority carriers.45 The irradiation of the plasmon coupler at various laser powers leaves the transfer characteristic qualitatively unchanged but shifted to higher gate voltages, implying an unchanged effective mobility (see inset of Fig. 1b and the corresponding mobility calculations in the supplementary information Figure S2). Notably, the diameter of the laser beam was only 1.3 µm so it was ensured that only the plasmon coupler was illuminated and no disturbing photocurrent is generated due to unintentional illumination of the adjacent Ge segment. The shift of the transfer curve is attributed to negatively charged traps inducing changes of the mobile charge carrier densities in the SBFET channel as is the case with photogating.46 However in contrast to common photogating, hot charge carriers occupying the traps are not generated via photo absorption but during the non-radiative decay of SPPs at the Al-Ge junction. To distinguish clearly, we will henceforth refer to this feedback effect as plasmongating.
Due to the same mode of action in analogy to photogating, the plasmon induced current IPl can be expressed by47
$${I}_{pl}\approx {g}_{m}\bullet \varDelta {V}_{th}$$
1
with \({g}_{m}\)the transconductance and \(\varDelta {V}_{th}\) the plasmon induced shift of the threshold voltage. With increasing laser power, the occupation of the limited number of traps saturates and thus the shift of the threshold voltage as shown in the inset of Fig. 1b. The transient measurements shown in Fig. 1c provide information on the relevant time constants for the filling and emptying of traps. When the plasmon coupler is illuminated for 3 s at VG = -40 V (VD = 100 mV) the rise and fall characteristic of the plasmon-induced current clearly shows that at least two mechanisms with very different time constants are present. These may be assigned to fast traps at the Ge/GeOx interface and slow traps in the oxide, respectively.48 Fast trap states directly located at the interface feature a short lifetime in the scale of microseconds which leads to the almost instantaneous increase of IPl at the beginning of the laser pulse. The slow states reside in the oxide layer and can be reached only by quantum mechanical tunneling resulting in long lifetimes even beyond minutes.48 Continuous filling of these slow traps leads to a further very slow increase of IPl during the 3s pulse. After the pulse, the emptying of the traps leads to an initially very fast drop in IPl, further gradually declining toward the initial current within a few seconds. As will be shown below, the temporal dynamics of the trapping processes depend on the operation mode (VG) of the SBFET and the intensity, duration and energy of the laser pulses, and thus on the SPP spike. The inset of Fig. 1c demonstrates an extreme example for the same SBFET at VG = 0 V. High energetic (λ = 532 nm) laser irradiation for 50 s leads to a decay time in the range of several minutes in accordance with trapping times of extreme slow traps in Ge oxide reported in previous studies.49,50 The temporal dynamics of the trapping processes and thus also the plasticity of the synaptic transistor also depend on the geometry of the SBFET. In order to investigate the influence of the Ge channel length but also of the SPP waveguides, devices were fabricated which have either the same waveguides (LWG = 3 µm in the inset of Fig. 1a) but different long Ge channels LGe or, conversely, Ge channels of the same length of LGe = 3 µm but SPP waveguides of different lengths.
The Al feed lines act as source/drain contacts of the SBFET and the part between the plasmon coupler and the Al-Ge junction simultaneously as a metal stripe SPPs waveguide of length LWG. The intensity of the SPPs propagating along the Al waveguide attenuates because of inevitable ohmic damping in metals,51,52 leak radiation53 and scattering at defects due to surface roughness.54,55 The SPP propagation length in the Al feed lines of the actual devices was determined to about 1.64 µm and 3.08 µm for an incident laser wavelength of λ = 532 nm and λ = 785 nm, respectively (see the supplementary information Figure S3). An optimized SPP waveguide and improved plasmon coupler would increase the overall current, the decisive factor for the dynamics of the device is however the geometry of the channel of the SBFET. Figure 2a shows the plasmon current as a function of the Ge channel length under laser irradiation with λ = 532 nm and λ = 785 nm at the same laser power of P = 500 µW at a drain bias of VD = 100 mV and VG = 0 V. In analogy to a common photodetector the plasmon induced current IPl follows56
with τl and τtr the effective lifetime of the SPP-induced carriers and transit time of the carriers through the transistor channel, respectively. The carrier transit time can be derived according to τtr = LGe2/(µ⸱VD) with µ the carrier mobility and VD the applied drain bias. Thus, the plasmon current appears to be inversely proportional to the square of the channel length as depicted in Fig. 2a. Notably, the effective carrier lifetime τl does not simply denote the lifetime of the injected hot carriers before recombination, but rather the lifetime of excess carriers due to trap induced plasmongating.57 This results in an overall higher plasmon current observed for green laser irradiation (λ = 532 nm). For Al, the band crossing close to the Fermi level near the W point allows interband transitions that originate from valence band states with energies ranging continuously from the Fermi level to the laser beam energy.58 Consequently, SPP decay results in hot electrons with a continuous energy distribution that extends from zero energy to the laser energy. For the green laser, more and higher energetic traps with longer lifetimes are filled up, increasing plasmongating and thus IPl. For the monolithic SBFET with a 3 µm long SPP waveguide and the shortest Ge channel of LGe = 600 nm, we calculated an external quantum efficiency of EQE = 2.5%. Although no optimization with respect to effective plasmon coupling or SPP waveguiding has been done yet, this is quite remarkable since for other electrical plasmon detectors EQEs of less than 1% have been reported.59,60 We think this is a direct consequence of the monolithic architecture of the device and the resulting improved momentum matching mentioned above.
Although there are many reports on the common photogating effect when a semiconductor is directly exposed to light, the particular mechanisms of SPP-induced hot carrier trapping are actually unexplored. An important difference is that photoabsorption produces hot electrons with the same energy. For plasmon induced electron injection from Al, these have a homogeneous energy distribution ranging from the fermi level to the energy of the excitation laser. Traps that get negatively charged by capturing these hot electrons may change both the carrier density\({\Delta }n\) as well as the carrier mobility \({\Delta }\mu\) and thus modify the conductivity Δσ of the SBFET according to:
with A a scaling parameter, P the light power and γ the dimensionless exponent of the power law.61 Fitting the data according to the power-current law we obtain a γ = 0.2 for a gate voltage of VG = -40 V. The sublinear relationship derives that the effective lifetime of excess carriers \({\tau }_{l}\) is inversely proportional to the injection rate G and is explained by the one-center recombination theory.62 While increasing the gate voltage in steps of 10 V the sublinear relationship changes gradually to superlinear (γ = 1.18) at VG = + 20 V. Accordingly, the SBFET becomes more sensitive to SPPs due to an increase of the excess carrier lifetime with increasing laser power. Such superlinear behavior is a typical phenomenon which occurs when two types of traps with different capture cross-sections are effective.63 The effectiveness and interaction of these two types of traps and how they determine the conductivity of the SBFET as a function of gate voltage are described in detail in the supplementary information (Figure S4).
2.3 Synaptic performance
To mimic the response of biological synapses featuring the impact of one neuron on another, the SBFET is regarded as a synapse with SPP spikes as the pre-synaptic stimulus, the source and drain electrodes as the post-synapse and the channel conductance represents the synaptic weight. Figure 3a shows the excitatory post-synaptic current (EPSC) when the plasmon coupler is irradiated by a 5.5 µs lasting λ = 785 nm laser pulse with a power of P = 8 µW at VG = -40 V. The generation of SPPs by a laser pulse is a non-linear process involving transfer of energy from the laser pulse to coherent oscillations of electrons in the SPP waveguide. The propagation of the SPPs along the metallic waveguide as well as the injection of hot electrons at the metal-semiconductor interface are known to be very fast processes on the ps time scale.64,65 Thus, a laser pulse applied to the plasmon coupler is practically equivalent to an instantaneous SPP spike at the Al-Ge interface of equal duration. The respective laser pulse generated SPP spike triggers an EPSC with a peak value of about 280 nA. When the laser illumination i.e. the SPP spike is off, the traps gradually empty, which leads to the typical EPSC decay recovering the initial dark current after about 25 µs. Figure 3b shows the peak values of EPSC and recovery time as a function of the SPP spike duration for a gate voltage of VG = 0 V and VG = -40 V. In both cases the maximum EPSC and the recovery time increase for short SPP spike duration and saturate for longer plasmon trigger. With increasing SPP spike duration, more and more charge trapping states are filled, increasing the plasmongating and thus the EPSC, until saturation when all available traps are filled in the time average. At VG = -40 V the EPSC is significantly higher since the gating effect of negatively charged traps is particularly effective in the p-type accumulation regime with holes as the majority carriers. For low laser power and high positive gate voltages with electrons as majorities, plasmongating even leads to a decrease in current comparable to negative photoconductivity (see the supplementary information Figure S5). For the recovery time, i.e. the time it takes for the EPSC to return to the base level after the end of the pulse, there is a very similar dependency with regard to the pulse duration (Fig. 3b). Initially increasing with the spike duration, it then tends to saturate at a few µs, the typical timescale for fast trap states located at the Ge/GeOx interface.48 Therefore, increasing the intensity (Fig. 1b) or the duration (Fig. 3b) of the SPP spikes increases both the amplitude of EPSC and the recovery time, simulating a transformation from short-term to long-term plasticity in biological systems.
A further parameter showing the short-term plasticity memory capacity of the synaptic SBFET is paired-pulse facilitation (PPF) which refers to the increases in the ratio between post-synaptic currents caused by two consecutive stimuli.66Fig. 3c demonstrates that the EPSC after the second SPP spike (A2) is greater than that after the first SPP spike (A1) induced with the same laser power of P = 8 µW and a pulse duration of 8 µs each, at an interval of ΔT = 5 µs. This typical PPF behavior is a consequence of the dynamics of the trap’s occupation by the SPP-induced hot electrons. If the second SPP spike is applied very close to the first one, not all traps will be emptied before the arrival of the second spike. Therefore, the number of filled traps will be larger than at the end of the first SPP spike and thus plasmongating more effective.
Figure 3d shows the quantitative PPF index:
$$PPF=100\%\times \frac{{A}_{2}-{A}_{1}}{{A}_{1}}$$
6
with A1 and A2 the peak values of EPSCs for the first and the second spike as a function of the interval time of the SPP spikes generated by λ = 785 nm laser pulses (P = 8 µW and duration time is 8 µs for each pulse). The PPF index of 19% at an interval time of ΔT = 2 µs decreases rapidly to ~ 7.0%, with the interval time increasing to 35 µs, due to effective detrapping with increasing interval time.
Within this work, we demonstrated the key functionality of a CMOS compatible transistor that response to plasmon signals in a neuron-/synapse‐like fashion. The plasmongating effect driven by SPP-induced hot carriers stimulates the SBFET performing typical synaptic behavior, such as the EPSC decay and PPF. Plasmon-stimulated synaptic transistors may rise as a new research field, which aims to transfer the well-known high-bandwidth of plasmonics to the area of neuromorphic computing. The aim of this work was foremost to investigate particular mechanisms of SPP-induced hot carrier trapping and to show the feasibility of a plasmon-stimulated synaptic transistor. The far-reaching goal of an on-chip neuromorphic platform may be realized using a monolithic structure covering an electrically-driven plasmon source, optimized SPP waveguides and high sensitivity synaptic transistors in a monolithic ultrascaled device. To control signals carried by only a few plasmons within only a few oscillation cycles may enable unprecedented circuit interconnectivity for the future of artificial intelligence in the big data era.
4.1 Device Fabrication:
Starting substrate was a GeOI wafer with a 75 nm-thick (100)-oriented Ge device layer, a 150 nm-thick buried SiO2 layer and a 500 µm-thick Si handle wafer. The wafer was cut into small pieces of 7 × 7 mm2 and immersed in buffered hydrofluoric acid (7:1) to remove the SiO2 capping layer, rinsed with deionization water and blow dried with nitrogen. Next, conventional photolithography and reactive ion etching (RIE) were conducted to pattern Ge microbars with two oval shaped trenches. Before the metallization, a 5 s HI (14%) dip was performed to remove the native oxide in order to guarantee intimate Ge/Al contacts. 120 nm-thick Al pads overlapping the Ge microbars were fabricated via photolithography, sputtering and lift-off techniques. Afterward, rapid thermal annealing at 674K initiated an Al-Ge exchange inducing the formation of monolithic Al-Ge-Al heterostructures. All details of the device processing are given in the supplementary information (Figure S1). For electrical and optical characterization, the sample was glued onto a copper plate using silver paste, enabling access to the handle wafer acting as a common back gate.
4.2 Electrical and optical characterization of SBFET:
A Keysight B1500A semiconductor analyzer was used to record all I/V and transfer curves. A µ-Raman setup (WITec Alpha 300) with a Zeiss 100X objective (NA = 0.75, WD = 4 mm) and a coupled CCD camera was utilized to focus the laser beam on the plasmon coupler. For light power-dependent measurements, a frequency doubled Nd:YAG laser (λ = 532 nm) and a near-infrared He/Ne laser (785 nm) were coupled into the µ-Raman system via optical fibers. The laser power was adjusted from 5 µW to 1 mW, measured by a calibrated Si photodiode.
4.3 Setup for synapse measurement:
A supercontinuum laser (SuperK Extreme, NKT) assisted with an NKT SuperK Select acoustic-optical tunable filter, ranging from 500 nm to 1600 nm, was used as the light source and coupled to the microscope via a polarization-maintaining optical fiber. To effectively excite SPPs the laser beam was linearly polarized perpendicular to the oval cutouts acting as plasmon coupler. We employed a function generator (Yokogawa FG300) to control the NKT laser output, generating laser pulses with the adjustable duration and pulse interval. The drain current was amplified by a Keithley 428 current amplifier and finally displayed and saved by a digital storage oscilloscope (Keysight DSOX1102G). All measurements were conducted at room temperature.
Author Declarations
Conflict of Interest
The authors have no conflicts to disclose.
Author Contributions
Z.S., L.W. and F.S. conducted the device fabrication. Z.S. performed the optoelectronic measurements. Z.S. and M.S. performed synapse measurements. A.L. conceived the project and contributed essentially to the experimental design. All authors analyzed the results and agreed on the manuscript.
Acknowledgments
The authors further thank the Center for Micro- and Nanostructures for providing the cleanroom facilities. Zehao Song acknowledges financial support from the China Scholarship Council (CSC. 201908420216).
- Sarah L. Harris David Money Harris. Digital Design and Computer Architecture ARM Edition. Elsevier (2016).
- Fortier, P. & Michel, H. Computer Systems Performance Evaluation and Prediction. Comput. Syst. Perform. Eval. Predict. 1–525 (2003) doi:10.1016/B978-1-55558-260-9.X5000-0.
- Clark, R. et al. Perspective: New process technologies required for future devices and scaling. APL Mater. 6, (2018).
- Forsell, M., Nikula, S., Roivainen, J., Leppänen, V. & Träff, J. L. Performance and programmability comparison of the thick control flow architecture and current multicore processors. J. Supercomput. 78, 3152–3183 (2022).
- Chen, A. A review of emerging non-volatile memory (NVM) technologies and applications. Solid. State. Electron. 125, 25–38 (2016).
- Carvalho, C. The gap between processor and memory speeds. Icca 27–34 (2002).
- Dai, G. et al. DIMMining: Pruning-Efficient and Parallel Graph Mining on Near-Memory-Computing. Proceedings - International Symposium on Computer Architecture vol. 1 (Association for Computing Machinery, 2022).
- Hajibaba, M. & Gorgin, S. A review on modern distributed computing paradigms: Cloud computing, jungle computing and fog computing. J. Comput. Inf. Technol. 22, 69–84 (2014).
- Beyond von Neumann. Nat. Nanotechnol. 15, 507 (2020).
- Brodowicz, M. & Sterling, T. A non von neumann continuum computer architecture for scalability beyond moore’s law. 2016 ACM Int. Conf. Comput. Front. - Proc. 335–338 (2016) doi:10.1145/2903150.2903486.
- Hey, T. Quantum computing: An introduction. Comput. Control Eng. J. 10, 105–112 (1999).
- Merolla, P. A. et al. A million spiking-neuron integrated circuit with a scalable communication network and interface. Science (80-.). 345, 668–673 (2014).
- Service, R. F. The brain chip. Science (80-.). 345, 614–616 (2014).
- Ohno, T. et al. Short-term plasticity and long-term potentiation mimicked in single inorganic synapses. Nat. Mater. 10, 591–595 (2011).
- Xu, W., Min, S. Y., Hwang, H. & Lee, T. W. Organic core-sheath nanowire artificial synapses with femtojoule energy consumption. Sci. Adv. 2, 1–8 (2016).
- Van De Burgt, Y. et al. A non-volatile organic electrochemical device as a low-voltage artificial synapse for neuromorphic computing. Nat. Mater. 16, 414–418 (2017).
- Li, E. et al. Synaptic Transistor Capable of Accelerated Learning Induced by Temperature-Facilitated Modulation of Synaptic Plasticity. ACS Appl. Mater. Interfaces 11, 46008–46016 (2019).
- Cho, S. W., Kwon, S. M., Kim, Y.-H. & Park, S. K. Recent Progress in Transistor-Based Optoelectronic Synapses: From Neuromorphic Computing to Artificial Sensory System. Adv. Intell. Syst. 3, 2000162 (2021).
- Chang, T., Jo, S. H. & Lu, W. Short-term memory to long-term memory transition in a nanoscale memristor. ACS Nano 5, 7669–7676 (2011).
- Wang, Z. Q. et al. Synaptic learning and memory functions achieved using oxygen ion migration/diffusion in an amorphous InGaZnO memristor. Adv. Funct. Mater. 22, 2759–2765 (2012).
- Kim, K., Chen, C. L., Truong, Q., Shen, A. M. & Chen, Y. A carbon nanotube synapse with dynamic logic and learning. Adv. Mater. 25, 1693–1698 (2013).
- Jo, S. H. et al. Nanoscale Memristor Device as Synapse in Neuromorphic Systems. 1297–1301 (2010) doi:10.1021/nl904092h.
- Li, H. K. et al. A light-stimulated synaptic transistor with synaptic plasticity and memory functions based on InGaZnOx-Al2O3 thin film structure. J. Appl. Phys. 119, (2016).
- Qin, S. et al. A light-stimulated synaptic device based on graphene hybrid phototransistor. 2D Mater. 4, (2017).
- Ahmed, T. et al. Optically Stimulated Artificial Synapse Based on Layered Black Phosphorus. Small 15, 1–12 (2019).
- Yang, Y., He, Y., Nie, S., Shi, Y. & Wan, Q. Light stimulated IGZO-based electric-double-layer transistors for photoelectric neuromorphic devices. IEEE Electron Device Lett. 39, 897–900 (2018).
- He, W. et al. A multi-input light-stimulated synaptic transistor for complex neuromorphic computing. J. Mater. Chem. C 7, 12523–12531 (2019).
- Feng, B. et al. Achieving Infrared Detection by All-Si Plasmonic Hot-Electron Detectors with High Detectivity. ACS Nano 13, 8433–8441 (2019).
- Salamin, Y. et al. 100 GHz Plasmonic Photodetector. ACS Photonics 5, 3291–3297 (2018).
- Ho, Y. L. et al. Plasmonic Hot-Carriers in Channel-Coupled Nanogap Structure for Metal-Semiconductor Barrier Modulation and Spectral-Selective Plasmonic Monitoring. ACS Photonics 5, 2617–2623 (2018).
- Gramotnev, D. K. & Bozhevolnyi, S. I. Plasmonics beyond the diffraction limit. Nat. Photonics 4, 83–91 (2010).
- Zia, R., Schuller, J. A., Chandran, A. & Brongersma, M. L. Plasmonics: the next chip-scale technology. Mater. Today 9, 20–27 (2006).
- Nikolajsen, T., Leosson, K. & Bozhevolnyi, S. I. In-line extinction modulator based on long-range surface plasmon polaritons. Opt. Commun. 244, 455–459 (2005).
- Liang, Y., Li, C., Huang, Y. Z. & Zhang, Q. Plasmonic nanolasers in on-chip light sources: Prospects and challenges. ACS Nano 14, 14375–14390 (2020).
- Berini, P. Surface plasmon photodetectors and their applications. Laser Photonics Rev. 8, 197–220 (2014).
- Dionne, J. A., Sweatlock, L. A., Atwater, H. A. & Polman, A. Plasmon slot waveguides: Towards chip-scale propagation with subwavelength-scale localization. Phys. Rev. B - Condens. Matter Mater. Phys. 73, 1–9 (2006).
- Knight, M. W. et al. Aluminum for plasmonics. ACS Nano 8, 834–840 (2014).
- Lee, D. N. & Lee, H. J. Effect of stresses on the evolution of annealing textures in Cu and Al interconnects. J. Electron. Mater. 32, 1012–1022 (2003).
- Wind, L. et al. Monolithic Metal-Semiconductor-Metal Heterostructures Enabling Next-Generation Germanium Nanodevices. ACS Appl. Mater. Interfaces 13, 12393–12399 (2021).
- Sistani, M. et al. Nanoscale aluminum plasmonic waveguide with monolithically integrated germanium detector. Appl. Phys. Lett. 115, (2019).
- Zhang, J., Zhang, L. & Xu, W. Surface plasmon polaritons: Physics and applications. J. Phys. D. Appl. Phys. 45, (2012).
- Schuck, P. J. Nanoimaging: Hot electrons go through the barrier. Nat. Nanotechnol. 8, 799–800 (2013).
- Knight, M. W. et al. Embedding plasmonic nanostructure diodes enhances hot electron emission. Nano Lett. 13, 1687–1692 (2013).
- Sistani, M., Staudinger, P. & Lugstein, A. Polarity Control in Ge Nanowires by Electronic Surface Doping. J. Phys. Chem. C 124, 19858–19863 (2020).
- Wang, D. et al. Surface chemistry and electrical properties of germanium nanowires. J. Am. Chem. Soc. (2004) doi:10.1021/ja047435x.
- Staudinger, P., Sistani, M., Greil, J., Bertagnolli, E. & Lugstein, A. Ultrascaled Germanium Nanowires for Highly Sensitive Photodetection at the Quantum Ballistic Limit. Nano Lett. 18, 5030–5035 (2018).
- Furchi, M. M., Polyushkin, D. K., Pospischil, A. & Mueller, T. Mechanisms of photoconductivity in atomically thin MoS2. Nano Lett. 14, 6165–6170 (2014).
- Kingston, R. H. Review of germanium surface phenomena. J. Appl. Phys. 27, 101–114 (1956).
- Ahn, Y. H. & Park, J. Efficient visible light detection using individual germanium nanowire field effect transistors. Appl. Phys. Lett. 91, 15–18 (2007).
- Cao, L. et al. Engineering light absorption in semiconductor nanowire devices. Nat. Mater. 8, 643–647 (2009).
- Liu, M., Pelton, M. & Guyot-Sionnest, P. Reduced damping of surface plasmons at low temperatures. Phys. Rev. B - Condens. Matter Mater. Phys. 79, 1–5 (2009).
- Wei, H. et al. Plasmon Waveguiding in Nanowires. Chem. Rev. 118, 2882–2926 (2018).
- Zhang, S. & Xu, H. Optimizing substrate-mediated plasmon coupling toward high-performance plasmonic nanowire waveguides. ACS Nano 6, 8128–8135 (2012).
- Kusar, P., Gruber, C., Hohenau, A. & Krenn, J. R. Measurement and reduction of damping in plasmonic nanowires. Nano Lett. 12, 661–665 (2012).
- Ditlbacher, H. et al. Silver nanowires as surface plasmon resonators. Phys. Rev. Lett. 95, 1–4 (2005).
- Sze, S. M., Li, Y. (Professor of electrical engineering) & Ng, K. K. Physics of semiconductor devices. 931.
- Kamieniecki, E. Effect of charge trapping on effective carrier lifetime in compound semiconductors: High resistivity CdZnTe. J. Appl. Phys. 116, 193702 (2014).
- Sundararaman, R., Narang, P., Jermyn, A. S., Goddard, W. A. & Atwater, H. A. Theoretical predictions for hot-carrier generation from surface plasmon decay. Nat. Commun. 5, 1–8 (2014).
- Sobhani, A. et al. Narrowband photodetection in the near-infrared with a plasmon-induced hot electron device. Nat. Commun. 4, 1–6 (2013).
- Qi, Z. et al. Au nanoparticle-decorated silicon pyramids for plasmon-enhanced hot electron near-infrared photodetection. Nanotechnology 28, (2017).
- Zhao, Q. et al. The role of traps in the photocurrent generation mechanism in thin InSe photodetectors. Mater. Horizons 7, 252–262 (2020).
- Bube, R. H. Photoelectronic properties of semiconductors. 318 (1992).
- Kao, K. C. 7 Electrical Conduction and Photoconduction PART I: E LECTRICAL C ONDUCTION. Dielectr. Phenom. Solids 503–504 (2004) doi:10.1016/B978-0-12-396561-5.50017-7.
- MacDonald, K. F., Sámson, Z. L., Stockman, M. I. & Zheludev, N. I. Ultrafast active plasmonics. Nat. Photonics 3, 55–58 (2009).
- Mongin, D. et al. Ultrafast photoinduced charge separation in metal-semiconductor nanohybrids. ACS Nano 6, 7034–7043 (2012).
- Li, B. et al. Mimicking synaptic functionality with an InAs nanowire phototransistor. Nanotechnology 29, (2018).
There is NO Competing Interest.
- SupplementaryInformationNC.docx
Supplementary Material See supplementary material for details of the device fabrication, the calculation of the field-effect mobility, SPP propagation length, nonradiative recombination theory and negative photoconductivity.