Circular polarization-resolved ultraviolet photonic artificial synapse based on chiral perovskite

Circularly polarized light (CPL) adds a unique dimension to optical information processing and communication. Integrating CPL sensitivity with light learning and memory in a photonic artificial synapse (PAS) device holds significant value for advanced neuromorphic vision systems. However, the development of such systems has been impeded by the scarcity of suitable CPL active optoelectronic materials. In this work, we employ a helical chiral perovskite hybrid combined with single-wall carbon nanotubes to achieve circularly polarized ultraviolet neuromorphic vision sensing and imaging. The heterostructure demonstrates long-term charge storage as evidenced by multiple-pulsed transient absorption measurements and highly sensitive circular polarization-dependent photodetection, thereby enabling efficient CPL-resolved synaptic and neuromorphic behaviors. Significantly, our PAS sensor arrays adeptly visualize, discriminate, and memorize distinct circularly polarized images with up to 93% recognition accuracy in spiking neural network simulations. These findings underscore the pivotal role of chiral perovskites in advancing PAS technology and circular polarization-enhanced ultraviolet neuromorphic vision systems.


List of Acronyms
The electron affinity is calculated by the equation: where hv is the energy of the He-I source (21.2 eV), W is the width of the UPS spectrum, and Eg is the bandgap of the H-PVK.
The direct bandgap of the H-PVK is calculated by the absorption-derivative method and the Tauc's equation as follow

𝐵 (𝐴ℎ𝑣) + 𝐸 = ℎ𝑣
where B1 is the constant, A is the absorbance, hv is the photo energy, and Eg is the bandgap.Supplementary Fig. 13 | Band alignments of H-PVK and SWNTs before and after connection.(Evac is the vacuum energy level, EF is the Fermi energy level).
Once H-PVK and SWNTs are connected, ground-state electron transfer occurs from PVK (with a low work function, higher EF, from UPS) to SWNTs (with a high work function, lower EF).This transfer continues until the Fermi levels align across the interface under equilibrium conditions, and results in interfacial energy band bending.Consequently, an electron depletion region can form at the H-PVK side, creating a potential barrier of 0.7 eV.This potential barrier is larger than the conduction band offset of 0.4 eV, making it unfavorable for photoexcited electron transfer from H-PVK to SWNTs.
Note that despite the fact that perovskite bandgap energy (including our chiral perovskites) is typically larger than that of SWNTs, no FRET energy transfer (requiring dipole-dipole interaction) or Dexter-like energy transfer (requiring wavefunction overlap) from perovskite film to SWNTs was reported.Since FRET or Dexter-like energy transfer requires the donor and accepter very close ~ 1-2 nm, which is usually occurs between quantum dot/SWNTs 1 .In our perovskite/SWNT heterostructure, the perovskite film is ~70 nm (Fig. S3), therefore, FRET or Dexter-like energy transfer from perovskite film to SWNTs should not be efficient.Supplementary Fig. 22 | Near-IR TA measurements to probe the photoexcited hole transfer to SWNTs.(a, c, e) TA spectra at various delay time after subtraction of the background signal and (b, d, f) normalized GSB dynamics probed at ~1050 nm for SWNTs, 1D-R/SWNTs, and1D-R/SWNTs, respectively.The excitation wavelength is 340 nm.The photobleaching (PB) band with peak at ~1050 nm arising from the stating filling of S11 exciton transition along with a fast decay lifetime (< 1ps) is observed in our SWNTs sample (Supplementary Figs.22a-b), which is consistent with previous results [1][2][3] .After addition of our chiral perovskites, an additional long-lived decay component probed at 1050 nm is observed in heterostructures (Supplementary Figs.22c-f), which arises from the stating filling of the separated holes.These observations are consistent with previous reports of the long-lived separated holes in SWNTs 1,4 .Note that the fast decay from S11 exciton is still existing in TA dynamics of heterostructures, which could be due to the energy transfer and/or the direct excitation of SWNTs by the pump laser.However, even if there are some short-lived excitons in SWNTs, they cannot contribute When the membrane potential value of one output neuron exceeds the threshold, the neuron fires and releases a spike to its next connections, and its membrane potential are reset.The fired neuron will also prevent other neurons from firing by lateral inhabitation.Within a period, the neuron keeps refractory state and cannot be fired.Moreover, the corresponding synapses that contributes to the firing result will be strengthened while synapses without contribution for the fire will be weakened.After the training, every category dataset is labeled with fixed threshold and training weights.In the test process, the images are input to the trained network, the output neurons are labeled to the pattern categories in terms of their most firing times to the corresponding input dataset.

Supplementary Fig. 1 |
Molecular structures of 1D-S and 1D-R chiral perovskites.Crystal structure of 1D-S and 1D-R from the view along b direction.In the molecular structure, the enantiomers (R-and S-α-MBA) remain in close proximity to the vertices of (PbI6) 4-octahedron, allowing for intense interactions between chiral organic and achiral inorganic parts.Supplementary Fig. 2 | Crystal characteristic of H-PVKs.(a), (b) XRD patterns of 1D-R and 1D-S films.The insets show the enlarged view of the XRD result range from 22° to 35°.XRD patterns of 1D-R and 1D-S films with the diffraction peaks of (002), (004), (006), and (008) suggest that the growth of spin-coated H-PVK features a preferred orientation along caxis.Supplementary Fig. 3 | Characterization of H-PVK films.Thickness of H-PVK film measured by surface profiler.The deposited H-PVK film has a thickness of approximately 70 nm.Supplementary Fig. 6 | Calculation of anisotropy factor of CD.The gCD spectra of 1D-R and 1D-S films.The dissymmetry of absorption can be calculated by the equation:  = CD[mdeg] 32980 × absorbance Supplementary Fig. 8 | Raman characteristic of SWNTs.Raman spectrum of SWNTs dip coating on SiO2/Si substrate.Clear observation of G bands (G + and G -) and D bands from Raman spectroscopy suggests the successful transfer of semiconducting SWNTs onto the Si substrate.Supplementary Fig. 9 | AFM characteristic of dip-coating SWNTs on SiO2/Si substrate.(a), (b) 2D and 3D AFM (5 × 5 μm) images of SWNTs network.c Extracted height data along white dash line in a.The calculated RMS from AFM measurements of the SWNTs surface is 1.2 nm, indicating a uniform SWNTs network is transferred onto the wafer with thickness of around 6 nm, indicating that a high-quality SWNTs network is obtained by the dip-coating method.Supplementary Fig. 12 | Band structure of H-PVK.(a) UPS spectra of H-PVK.(b) The bandgap of H-PVK calculated by employing Tauc's plot of (αhv) 2 versus energy.
to the photocurrent due to the fast recombination, it thus could only affect the quantum efficiency of device, and won't affect the fundamental concept and function of our LCP/RCP resolvable artificial synapse devices.Supplementary Fig. 23 | Laser spike-numbers dependent change of band edge PB amplitudes probed at 375 nm and 5 ns of 1D-S (upper panel) and 1D-S/SWNTs (lower panel) under 340 nm LCP/RCP excitation.Supplementary Fig. 29 | Flow diagram of the implementation of the SNNs.Each dataset is encoded as a spike train that conforms to the Poisson equation, and then input to the neurons.The fire rate of neurons relies on the intensity of the corresponding pixel in the dataset image.For an MNIST image, if the spiking probability of a white pixel is 100%, and a black pixel never generates a spike.In the training process, the membrane potential of each output neuron accumulates by the input spikes and weight from other connecting synapses.