Page 73 - Read Online

Na et al. Soft Sci. 2026, 6, 25 Page 5 of 22

Schottky junction
A Schottky junction forms when a metal and a semiconductor with different work functions come into
contact, inducing charge transfer and Fermi-level alignment at equilibrium [46,47] . This generates an interfacial
potential barrier and a built-in electric field, bending the semiconductor energy bands [Figure 2Ai]. Upon
illumination, photogenerated holes are driven toward the metal by the built-in potential, while electrons are
blocked by the interfacial barrier from moving toward the metal and instead diffuse into the semiconductor
bulk [Figure 2Aii]. Meanwhile, a fraction of the photogenerated carriers can be trapped at interfacial or
defect sites, and their gradual detrapping after illumination leads to persistent photocurrent relaxation
[Figure 2Aiii]. These time-dependent trapping and release processes give rise to synaptic potentiation,
depression, and relaxation under zero-power operation, mimicking biological short- and long-term
plasticity [32-36] . Based on the intrinsic built-in electric field and slow decay, Schottky junction-based
zero-power optoelectronic synapses can operate without external electrical bias. As a result, synaptic
functions such as excitatory postsynaptic current (EPSC), paired-pulse facilitation (PPF), and short-term to
long-term memory (STM-to-LTM) transition can be realized without electrical power consumption. These
works highlight the versatility of Schottky junctions as a foundational mechanism for zero-power
optoelectronic synapses. In this section, we review representative Schottky junction-based zero-power
optoelectronic synapses operating without external electrical bias, focusing on their operating principles,
synaptic functionalities, and applications.

Ren et al. recently demonstrated a planar inorganic Schottky synapse based on single-crystalline Bi O Se
2
2
nanosheets containing selenium vacancies (Bi O Se-V ), synthesized via physical vapor deposition [Figure
2
Se
2
2B] . The vacancy-induced trap states, together with the Schottky contact, enabled operation across a broad
[32]
spectral range, from ultraviolet (UV) to near-infrared, without external electrical bias [Figure 2C]. The device
exhibited typical synaptic characteristics, including EPSC, PPF, and STM-to-LTM transition, and further
demonstrated frequency-dependent plasticity for high-pass optical filtering [Figure 2D].
To expand both spectral responsiveness and circuit-level adaptability, Dai et al. developed a zero-power
retinomorphic photodiode array based on a Schottky junction between poly(3-hexylthiophene-2,5-diyl)
(P3HT) and Al [Figure 2E] . The strong visible-light absorption of P3HT and defect-induced photogating
[33]
allowed inhibitory postsynaptic current (IPSC) and paired-pulse depression (PPD), demonstrating
reconfigurable learning behavior without external electrical bias [Figure 2F]. When integrated into a 10 × 5
array, the system successfully reproduced retina-inspired photopic adaptation, emulating the attenuation of
signal intensity under prolonged illumination [Figure 2G].

To approach optical functionality of in-sensor reservoir computing, Lao et al. developed an ultralow-power
machine vision system based on a zero-power sensor reservoir, exploiting a Schottky junction between the
Cs AgBiBr photosensitive layer and Au electrode [Figure 2H] . By embedding a ferroelectric-induced
[34]
6
2
potential well at the shoulder of the Schottky barrier, the device significantly prolonged the dwell time of
photogenerated carriers, thereby enabling nonlinear EPSC coupling under pulsed optical stimulation [Figure
2I]. This potential well-assisted carrier retention produces the spatiotemporally correlated EPSC responses
essential for reservoir computing. The device exhibits key synaptic functions including PPF,
spike-rate-dependent plasticity (SRDP) and spike-number-dependent plasticity (SNDP), demonstrating
strongly tunable synaptic plasticity. In addition, the evolution of the 25 × 4 memristor conductance map
from an initially random weight distribution to a well-trained configuration optimized for vehicle-flow
classification [Figure 2J].