Electro-Photo-Sensitive Memristor for Neuromorphic and Arithmetic Computing

Abstract

We present optically and electrically tunable conductance modifications of a site-controlled quantum dot memristor. The conductance of the device is tuned by electron localization on a quantum dot. The control of the conductance with voltage and low power light pulses enables applications in neuromorphic and arithmetic computing. As in neural networks, applying pre-and post-synaptic voltage pulses to the memristor allows to increase (potentiation) or decrease (depression) the conductance by tuning the time difference between the electrical pulses. Exploiting state-dependent thresholds for potentiation and depression, we were able to demonstrate a memory-dependent induction of learning. The discharging of the quantum dot can further be induced by low power light pulses in the nW-range. In combination with the state-dependent threshold voltage for discharging, this enables applications as generic building blocks to perform arithmetic operations in bases ranging from binary to decimal with low power optical excitation. Our findings allow the realization of optoelectronic memristor-based synapses in artificial neural networks with a memory-dependent induction of learning and enhanced functionality by performing arithmetic operations.

Mimicking of pulse shape-dependent learning rules with a quantum dot memristor

“…38 For voltage differences between the two terminals ΔV = Vpr -Vpo that exceed the threshold voltages for charging Vc ≈ -1.9 V and discharging Vd ≈ 3.9 V, the amount of charges is raised and lowered, respectively. 39 The switching between high and low conductances (see Fig. 1(b)) is comparable to other memristor realizations, e.g.…”

“…58 In Ref. 39, the relative conductance change ΔG/G0 of the present device for the asymmetric Hebbian learning rule was found to be independent on G0 for depression, but shows a maximum at medium G0 conductance values for potentiation. A dependency of the learning rules on the initial conductance was also presented for an Al2O3/TiO2-x memristor in Ref.…”

“…In contrast to the previous proposals, the presented device is based on the mature III-V semiconductor platform that enables optical conductance control with low power light pulses. 39 Thus the memductance state can be controlled either by optical or electrical pulses or by combinations of both, which allows the integration with photodetectors as sensory neurons. The conductance control is further sensitive to the wavelength of incoming light 51 , which was also demonstrated with other memristors 52 and memcapacitors 53 and enables encoding information in the wavelength.…”

“…The two discharging regimes (above and below Vd) have different timescales and are beneficial to perform arithmetic operations in tunable bases with more gradual conductance changes occurring at voltage pulse slightly below Vd. 39 Fig. 1(c) shows the voltage pulses that are required and used to emulate the four learning rules and applied to the drain (red) and source (blue) contacts.…”

“…1(d), the voltage difference for the considered pulses exceeds Vd for positive time difference leading to the discharging. 39 For negative time differences, |ΔV| exceeds |Vc|. Fig.…”