Page 75 - Read Online

Na et al. Soft Sci. 2026, 6, 25 Page 7 of 22

Heterojunction
Unlike Schottky junctions, heterojunctions form at the interface between two semiconductors with different
band structures. Fermi-level alignment through charge transfer creates both conduction- and valence-band
offsets, generating a built-in electric field that extends across the depletion region [Figure 3Ai] [47,48] . Under
illumination, excitons generated in the active layer dissociate only when they reach this interface, from which
photogenerated electrons and holes are efficiently separated and driven in opposite directions, producing a
photocurrent without external electrical bias [Figure 3Aii]. Concurrently, a portion of carriers becomes
trapped at interfacial or defect-related states, and their gradual release after light removal induces persistent
photocurrent relaxation, mimicking synaptic potentiation and depression under zero-power operation
[Figure 3Aiii] [37-42] . Compared with Schottky systems, heterojunctions offer more tunable band alignment and
enhanced charge separation efficiency, enabling broader spectral sensitivity and more versatile synaptic
behaviors. These characteristics make heterojunction-based zero-power optoelectronic synapses particularly
attractive for zero-power operation, as synaptic modulation can be governed by interfacial band engineering.
By carefully selecting semiconductor pairs with complementary energy levels, the built-in electric field and
carrier transport pathways can be precisely controlled, modulating synaptic weight, retention time, and
response speed. We review representative heterojunction-based optoelectronic synapses, highlighting
operating principles, synaptic functionalities, and applications.

Zheng et al. reported a retina-inspired optoelectronic synapse based on a vertical GaN/BiFeO 3
heterojunction [Figure 3B] . The ferroelectric BiFeO layer, containing abundant oxygen vacancies, acted as
[37]
3
both a UV absorber and a charge-trapping reservoir. The built-in photovoltaic field across the
heterojunction enabled operation without external electrical bias under ultra-weak UV illumination. The
device exhibited a wide range of synaptic plasticity, including EPSC, PPF, STM-to-LTM transition, and
learning-forgetting-relearning behavior, with a maximum PPF index of 200% and a minimum detectable UV
intensity of 0.68 μW·cm [Figure 3C]. UV-modulated negative polarization and vacancy ionization enhanced
-2
persistent photoconductivity, allowing the synaptic weight to be modulated by invisible UV light and used
for pattern recognition, including letter images [Figure 3D].

Beyond inorganic systems, Hao et al. developed zero-power artificial optoelectronic synapses based on
ultrathin organic asymmetric C -BTBT (dioctylbenzothienobenzothiophene)/F CuPc (copper
8
16
hexadecafluorophthalocyanine) heterojunction films [Figure 3E] . The efficient conversion of UV light into
[38]
photocurrent enabled zero-bias synaptic functions such as EPSC, PPF, and SRDP without external electrical
bias. Specifically, the SRDP behavior allows high-pass filtering for image preprocessing with a cutoff
frequency of 6.9 Hz [Figure 3F]. This high-pass filtering effect was further utilized for image sharpening,
where slowly varying background components were suppressed while high-frequency edge features were
selectively enhanced, enabling more efficient visual recognition [Figure 3G]. By tuning gate voltage to
modulate interfacial charge transfer, the platform extended its response from UV to the near-infrared region,
achieving multispectral sensing and expanding the perceptual range beyond that of the biological retina.

Cheng et al. demonstrated a zero-power optoelectronic memristor based on a ZnO/WO heterojunction
X
[Figure 3H], capable of mimicking key retinal adaptation phenomena, including desensitization and Weber’s
law . Under pulsed optical stimulation without an external electrical bias, the photovoltaic effect, coupled
[39]
with charge trapping at the heterojunction, produced transient postsynaptic currents with gradual decay,
analogous to photoreceptor adaptation [Figure 3I]. The device further exhibited background-dependent
response to flash stimuli, representing Weber’s law [Figure 3J]. By integrating the memristor with a
NbO -based threshold-switching element, the system converted continuous light inputs into dynamic spike
X
trains, enabling adaptive image preprocessing and overexposed image recognition with high accuracy [Figure
3K and L].