Researchers have engineered a monolithic nonlinear back-to-back photodiode that achieves filter-free spectral discrimination between visible and near-infrared light within a single photodetector pixel by altering the applied electrical bias. Conventional optical sensing platforms require external bandpass filters, complex optical assemblies, or multi-pixel arrays to resolve distinct wavelength bands, introducing substantial optical attenuation and complicating integration. By integrating a methylammonium lead iodide perovskite absorber with an underlying silicon substrate through a conductive polymeric interlayer, the researchers generated distinct electrical fingerprints from superimposed visible and near-infrared illumination, enabling simultaneous two-color decoding directly from a single electrical output terminal without external optical filters.
The development addresses longstanding engineering constraints inherent to standard silicon photodiodes and traditional vertically stacked dual-band photodetector designs. Conventional silicon exhibits broad, nonselective absorption across visible and near-infrared spectra, forcing commercial imaging systems to incorporate discrete optical dye or interference filters that exhibit vulnerability to incidence angles and cause optical losses. Previous attempts to realize filter-free wavelength selectivity relied on vertically stacked back-to-back photodiodes with opposing junction polarities or stacked organic photodetector junctions. In those serial configurations, applying a specific bias polarity preferentially activates charge extraction in only one junction while the oppositely oriented junction acts as a passive series resistor. This substantial internal resistance impedes carrier transport, depresses responsivity, degrades spectral separation, and restricts linear dynamic range unless driven at potentials exceeding one volt. High driving voltages inevitably amplify dark current and electrical noise, establishing an unfavorable trade-off between driving power, operational efficiency, and signal fidelity in compact multi-wavelength sensors.
To resolve these resistive bottlenecks under sub-volt operating conditions, the team constructed a monolithic heterostructure incorporating a solution-processed high-conductivity polymer interlayer between silicon and perovskite. The device architecture comprises an eight-square-millimeter active area fabricated on an n-type silicon substrate approximately 500 micrometers thick, topped sequentially with a 50-nanometer layer of high-conductivity PEDOT:PSS, a 100-nanometer methylammonium lead iodide perovskite absorber, a 50-nanometer phenyl-C61-butyric acid methyl ester layer, and an evaporated gold and silver grid electrode measuring 65 nanometers in total thickness. The critical modification involves using high-conductivity formulation PH1000, possessing a PEDOT-to-PSS weight ratio of one to 2.5 and a work function of 4.92 electron-volts, rather than standard low-conductivity formulation AI4083, which contains a one to six ratio and exhibits a 5.09 electron-volt work function. The acidic interactions between the polymer and the perovskite precursor induce localized, defect-assisted interfacial pathways that optoelectronic designs typically avoid, which the researchers intentionally harnessed to lower junction resistance and facilitate cooperative carrier transport across both photoactive regions.
Optical design simulations and cross-sectional structural analyses confirmed how layer dimensions govern wavelength-selective carrier generation within the vertical device stack. High-angle annular dark-field scanning transmission electron microscopy combined with energy-dispersive X-ray spectroscopy confirmed uniform layer deposition and distinct elemental localization of lead, iodine, carbon, and silicon across the heterostructure interfaces. Transfer matrix simulations across the 400 to 900 nanometer spectrum revealed that visible photons are predominantly absorbed within the top perovskite layer and electron-transport layer due to the 1.59 electron-volt optical bandgap of methylammonium lead iodide. In contrast, near-infrared wavelengths penetrate deeper into the underlying silicon substrate, creating complementary spatial absorption profiles. By systematically modeling perovskite film thicknesses from 50 to 350 nanometers, the researchers determined that an optimal thickness of 100 nanometers absorbs sufficient visible light to produce strong visible photocurrent while allowing adequate transmission to excite the underlying silicon, thereby establishing balanced dual-absorber photoactivation necessary for bias-tunable electrical operation.
Current-voltage characterization under calibrated monochromatic illumination demonstrated dramatic reconfigurability across a modest operating window spanning minus 0.5 volts to positive 0.5 volts. Under negative 0.5-volt bias, both visible light at 625 nanometers and near-infrared light at 850 nanometers produced linear photocurrent responses exceeding dark current, yielding linear dynamic ranges of at least 100 decibels with power-law linearity slopes of 0.97 and 0.96, respectively. At zero volts, both wavelength channels exhibited comparable photocurrent amplitudes without spectral selectivity, maintaining linear dynamic ranges of approximately 60 decibels. When the bias shifted to positive 0.5 volts, the near-infrared response remained monotonic across irradiance levels, whereas the visible-light response displayed pronounced nonlinear behavior. Under weak visible illumination below 100 microwatts per square centimeter, visible photocurrent was suppressed below the dark current level, generating a negative polarity. As incident irradiance exceeded that specific 100-microwatt threshold, the visible response underwent an abrupt polarity reversal to positive current, establishing an irradiance-dependent electrical fingerprint unique to visible excitation.
The authors explained this distinctive behavior by analyzing competing photovoltaic and photoconductive mechanisms that govern carrier separation under opposing junction orientations. Under near-infrared illumination, photons penetrate deep into silicon, exciting photovoltaic carrier separation across the reverse-biased silicon-polymer junction alongside bulk photoconductive enhancements. At positive bias, the bulk photoconductive current flows in the same direction as the dark current, effectively compensating for the diminished photovoltaic contribution and maintaining a strictly positive, monotonic output. Conversely, visible illumination excites competing, oppositely oriented photovoltaic currents at both the silicon-polymer and perovskite-polymer interfaces. At positive 0.5 volts, the residual built-in field at the silicon junction drives reverse photocurrent opposing forward perovskite photocurrent. Because interfacial defect states accelerate carrier recombination at low visible intensities, the perovskite photovoltaic current remains suppressed, allowing silicon-side current to dominate. Once visible irradiance exceeds 100 microwatts per square centimeter, intense carrier generation overcomes recombination bottlenecks, causing perovskite photocurrent to dominate and reversing overall polarity.
Experimental validation using five systematically modified control architectures confirmed that dual photoactive absorbers and the high-conductivity polymer interlayer are strictly essential for spectral selectivity. A device fabricated without any polymer interlayer exhibited negligible carrier extraction at negative 0.5 volts, confirming that photogenerated carriers could not overcome the resistive perovskite barrier. A control incorporating a standard low-conductivity polymer layer produced photocurrents two orders of magnitude lower, measuring only 0.1 microamperes at negative 0.5 volts, mirroring results from a bilayer control device where perovskite contacted low-conductivity polymer. Perovskite-free silicon control devices exhibited broadband photoresponses lacking spectral discrimination, while replacing silicon with transparent fluorine-doped tin oxide on glass eliminated near-infrared sensitivity. Comparative evaluations confirmed that only the monolithic combination of silicon, high-conductivity polymer, and optimized perovskite produces the deliberate interfacial leakage pathways required to reconfigure photoresponse linearity and unlock bias-controlled dual-band discrimination within a sub-volt potential range.
Dynamic performance characterization demonstrated robust temporal response metrics, high sensitivity, and stable operational endurance under cyclic pulsed illumination. Shot-noise-limited calculations yielded peak specific detectivities of 6.4 times ten to the eleventh Jones at 625 nanometers and 1.1 times ten to the twelfth Jones at 850 nanometers under minus 0.5 volts. Under pulsed light excitation, the device achieved rapid response dynamics, exhibiting rise and decay times of 5.1 and 12 microseconds for visible light and 2.8 and 99 microseconds for near-infrared light at negative 0.5 volts, alongside near-infrared rise and decay times of 120 and 99 microseconds at positive 0.5 volts. Operational repeatability testing across 100 consecutive light-switching cycles over 500 seconds demonstrated stable photocurrent generation without discernible signal drift or structural degradation. To assess long-term operational stability, the photoresponse of the device under visible and near-infrared illumination was monitored for 10 days.
To demonstrate computational demultiplexing of mixed optical signals, the investigators evaluated four supervised machine learning algorithms using experimental multi-bias photocurrent databases. When training Ridge regression, extreme gradient boosting, multilayer perceptron, and Gaussian process models on two bias states at minus 0.5 volts and zero volts, all algorithms exhibited substantial scatter, with the best-performing multilayer perceptron yielding an average determination coefficient of 0.7483 and mean absolute error of 5.53 microwatts per square centimeter. Incorporating the positive 0.5-volt photocurrent response as a third input feature clustered predictions tightly along the parity line across all four algorithms. The three-bias Gaussian process regression model achieved extraordinary prediction fidelity, demonstrating coefficients of determination of 0.9993 for the visible channel and 0.9999 for the near-infrared channel, corresponding to an average coefficient of 0.9996 and an average mean absolute error of 0.20 microwatts per square centimeter, representing an approximate 28-fold accuracy improvement over two-bias inputs.
The research team applied the bias-programmable detector to real-world optical monitoring tasks, demonstrating transmission pulse oximetry and agricultural reflectance diagnostics. In human volunteer trials using simultaneous red and near-infrared illumination through an index finger, the sensor recorded stable photoplethysmography waveforms across all three biases, resolving pulse rates of 86, 80, and 73 beats per minute that matched commercial reference ranges between 70 and 92 beats per minute. Calculating peripheral oxygen saturation via the modified Beer-Lambert law yielded 95.4 percent, comparable to the 98 to 100 percent range measured by a clinical fingertip oximeter, although the authors noted that tissue scattering, variable optical path lengths, and uncalibrated empirical coefficients introduce measurement uncertainties requiring further clinical calibration. In agricultural tests evaluating 30 leaves across healthy, stressed, and dead conditions from a campus garden, a Gaussian process classifier trained on three-bias reflectance photocurrents correctly identified 29 of 30 samples, achieving over 96 percent accuracy in normalized difference vegetation index classification despite limitations regarding single-species sampling.
Subject of Research: Bias-programmable single-pixel photodetector for filter-free visible and near-infrared spectral discrimination
Article Title: Bias-programmable linear-to-nonlinear photoresponse for filter-free single-pixel VIS–NIR discrimination
Article References: Shin, J., Alam, M., Kim, J., Lee, Y. G., Jeon, M. K., Kim, H., & Jung, G. Y. (2026). Bias-programmable linear-to-nonlinear photoresponse for filter-free single-pixel VIS–NIR discrimination. Nano Convergence, 13(1), Article 52. https://doi.org/10.1186/s40580-026-00585-7
Image Credits: AI Generated
DOI: 10.1186/s40580-026-00585-7
Keywords: photodiode, perovskite, silicon photodetector, spectral discrimination, near-infrared, visible light, PEDOT:PSS, linear dynamic range, pulse oximetry, normalized difference vegetation index, machine learning, back-to-back photodiode, optical sensing, thin films, photoconductive effect, photovoltaic effect, multispectral imaging, optoelectronics
Cite Scienmag News
Teresa Odom. (November 16, 2026). Single Photodiode Pixel Separates Visible and Infrared Light Using Programmable Voltage. Scienmag. https://scienmag.com/single-photodiode-pixel-separates-visible-and-infrared-light-using-programmable-voltage/
Teresa Odom. "Single Photodiode Pixel Separates Visible and Infrared Light Using Programmable Voltage." Scienmag, 16 November 2026, https://scienmag.com/single-photodiode-pixel-separates-visible-and-infrared-light-using-programmable-voltage/. Accessed 16 November 2026.
Teresa Odom. "Single Photodiode Pixel Separates Visible and Infrared Light Using Programmable Voltage." Scienmag. November 16, 2026. https://scienmag.com/single-photodiode-pixel-separates-visible-and-infrared-light-using-programmable-voltage/








