All-in-plane image sensors free from readout integrated circuits

This paper introduces and experimentally validates a novel all-in-plane image sensor architecture that eliminates the need for individual pixel readout circuits by using neighbor-connected photoresistive pixels and reconstructing images via electrical impedance tomography, thereby enabling unprecedented simplification for ultra-miniaturized and heterogeneously integrated detectors.

The Gist

Imagine you are trying to figure out where a fire is burning inside a large, dark warehouse.

The Old Way (Traditional Sensors):
In a standard camera or night-vision device, the warehouse is divided into thousands of tiny, individual rooms (pixels). Inside every single room, there is a tiny security guard (a transistor) and a dedicated telephone wire running from that room all the way to the control room. If a fire starts in Room 42, the guard in Room 42 picks up the phone, and the control room knows exactly where the fire is.

• The Problem: Building a warehouse with millions of rooms, each with its own phone wire and guard, is incredibly expensive, difficult to build, and impossible to do with some new, fragile materials (like graphene or special oxides) because you can't easily wire them up.

The New Way (This Paper's Innovation):
The researchers in this paper came up with a clever trick. They removed all the individual guards and all the telephone wires.

Instead, they built a giant, flat, conductive floor (the sensor) made of a special material that changes its electrical resistance when light hits it. They only put a few "phones" around the very edge of the warehouse floor.

Here is how they find the fire (the light) without looking inside:

1. The "Current" Game: They send a small electrical current into the floor from one corner (like pouring water into a sponge).
2. The "Listening" Game: They measure the voltage (electrical pressure) at all the other corners around the edge.
3. The Magic: If a light hits a specific spot on the floor, it changes the "texture" of the floor at that spot, making it harder for the electricity to flow through. This changes the pattern of voltage they measure at the edges.
4. The Detective Work: By sending currents from different corners and listening at different edges, they get a massive amount of data. A computer then uses a mathematical recipe (called Electrical Impedance Tomography) to work backward. It's like a detective looking at the ripples on a pond to figure out exactly where a stone was dropped, even though they can't see the stone itself.

Creative Analogies to Help You Visualize

1. The "Sponge" Analogy
Imagine a giant, flat sponge. If you squeeze one corner, water squirts out the other sides.

• Traditional Sensor: You have a tiny tube attached to every square inch of the sponge to measure exactly how wet each spot is.
• This Sensor: You only have tubes at the four edges. You squeeze different corners and watch how the water flows out the edges. If you put a heavy rock (light) on a specific spot in the middle, it changes how the water flows. By analyzing the flow patterns at the edges, a computer can draw a map of exactly where the rock is, even though it never touched the middle of the sponge.

2. The "Orchestra" Analogy
Think of the sensor as a giant drum skin.

• Traditional Sensor: Every inch of the drum has a microphone attached to it.
• This Sensor: You only have microphones at the very edge of the drum. You tap the drum in different places (injecting current). If someone hits the drum with a mallet (light) in the middle, it changes the sound of the vibrations traveling to the edge. By listening to the "sound" (voltage) at the edge while tapping different spots, the computer can reconstruct the image of where the mallet hit.

Why Is This a Big Deal?

• Simplicity: You don't need complex wiring inside the sensor. This makes the device much cheaper and easier to build.
• New Materials: Because you don't need to wire up every single tiny pixel, you can use "weird" new materials (like graphene or vanadium oxide) that are great at sensing light but are too fragile or difficult to wire up using traditional methods.
• Scalability: You can make these sensors huge without needing millions of wires. The number of wires needed only grows with the square root of the size, not the total size. It's like needing only 20 wires to read a 264-pixel sensor, whereas a traditional one would need 264 wires.

The Results

The team tested this with two different materials:

1. Graphene: A super-thin, super-strong carbon material. They built a small 24-pixel sensor and successfully reconstructed images of where a laser was shining.
2. Vanadium Oxide: A material often used in thermal cameras. They built a larger 264-pixel sensor and again successfully found the "hot spots" of light.

In a nutshell: They turned a complex, hard-to-wire electronic puzzle into a simple "guess the location" game played with electricity and math. This opens the door to cheaper, easier-to-make cameras for everything from night vision to medical imaging, using materials that were previously too difficult to use.

Technical Summary

Here is a detailed technical summary of the paper "All-in-plane image sensors free from readout integrated circuits."

1. Problem Statement

Conventional high-resolution matrix image sensors rely on Readout Integrated Circuits (ROICs) that require individual electrical access to every pixel. This architecture presents significant challenges:

• Integration Complexity: Heterogeneous integration of photosensitive materials (e.g., graphene, vanadium oxide, perovskites) with silicon ROICs requires ultra-precise wafer alignment, which is expensive and yield-limiting.
• Fabrication Barriers: Monolithic integration of amplifying transistors within each pixel involves complex lithography, doping, and high-temperature annealing.
• Scalability Limits: For novel optoelectronic materials, the lack of established integration processes limits research-grade imagers to very small pixel counts (tens of pixels), preventing systematic scaling studies.
• Crossbar Limitations: Alternative crossbar architectures suffer from signal crosstalk, parasitic capacitance, and are generally restricted to diode-type detectors, which are scarce for long-wavelength radiation.

The authors aim to eliminate the need for vertical integration of light-sensitive arrays with readout electronics, enabling imaging with simplified architecture and diverse materials.

2. Methodology: In-Plane Tomographic Imaging

The proposed solution is an all-in-plane image sensor that utilizes Electrical Impedance Tomography (EIT) principles to reconstruct images without individual pixel addressing.

• Sensor Architecture:
  • The sensor consists of a rectangular lattice of photoresistive pixels connected neighbor-to-neighbor in a single plane.
  • There are no internal readout circuits; electrical contacts are located only at the perimeter of the matrix.
• Operating Principle:
  • Current Injection: A bias current is sequentially injected between different pairs of perimeter contacts.
  • Voltage Measurement: The resulting photovoltage is measured between all other pairs of perimeter contacts.
  • Data Acquisition: This process generates a dataset of non-local photoresistances. The spatial distribution of light intensity is encoded in the potential distribution across the entire network.
• Mathematical Framework:
  • The relationship between local photoresistance changes ($\delta\rho$) and boundary photovoltages ($\delta V$) is linearized.
  • The system is modeled as a linear equation: $\hat{S} \delta\rho = \delta\rho_{eff}$, where $\hat{S}$ is a dimensionless sensitivity matrix derived from Kirchhoff simulations or absolute EIT calibration.
  • Reconstruction: The image is reconstructed by solving this linear system using Non-Negative Least Squares (NNLS) to ensure physical meaningfulness (resistivity changes are of uniform sign).
• Key Physical Insight:
  • The method relies on the reciprocity of current injection and voltage measurement (Tellegen's theorem).
  • Unlike conventional EIT used for conductivity mapping (which is non-linear and ill-posed), this application is a linear problem because the photoresistance perturbation is small compared to the DC resistance, ensuring mathematical stability and low computational cost.

3. Key Contributions

• Novel Architecture: Introduction of a "readout-free" imaging paradigm where optical information is encoded collectively in the resistive response of a single planar sheet, removing the need for transistor integration and vertical metallization.
• Material Agnosticism: The method is independent of the microscopic mechanism of photoresistance (e.g., bolometric heating, interband excitation, or photogating), making it applicable to a wide range of emerging materials (2D materials, bulk semiconductors).
• Scalability: The number of required electrical contacts ($N_e$) scales as the square root of the total pixel count ($N_e \approx \sqrt{2N_{pix}}$). This allows large-format matrices to be read out through a limited number of connections.
• Algorithmic Stability: The reconstruction is mathematically stable and robust against variations in pixel resistivity and photosensitivity, provided a correction procedure (based on translational symmetry) is applied.

4. Experimental Results

The authors validated the concept using two distinct mid-infrared ($\lambda_0 = 8.6 \, \mu m$) detector matrices:

A. Multilayer Graphene (MLG) Sensor (24 Pixels)

• Setup: $3 \times 3$ pixel matrix (24 photoresistors total) with 12 perimeter contacts.
• Performance: Successfully reconstructed images of a focused laser spot in various corners.
• Correction: Implemented a translation-symmetry-based correction to account for non-uniform pixel responsivity (observed ~40% RMS scatter). After correction, the reconstructed spot position matched the nominal illumination location with minimal cross-talk.

B. Amorphous Vanadium Oxide (VOx) Sensor (264 Pixels)

• Setup: $11 \times 11$ pixel matrix (264 photoresistors) with 20 perimeter contacts.
• Challenge: The number of independent measurements (190) was less than the number of pixels (264), creating an underdetermined system.
• Solution: The authors used a reduced basis of smooth Gaussian functions to expand the photoresistivity change, effectively clustering resistors to solve the inverse problem.
• Performance: Successfully localized illumination spots near edges and the center of the matrix. While some artifacts (secondary spots) appeared due to limited contacts, the primary spot position was identified stably.

5. Significance and Future Outlook

• Technological Simplification: This approach offers a fabrication-tolerant alternative to CMOS-based matrices, potentially lowering costs and enabling the use of materials that are difficult to integrate with silicon ROICs.
• Fundamental Research Tool: It enables the creation of large-scale matrix imagers for studying new optoelectronic materials without the bottleneck of custom electrode routing.
• "Microscopy without a Microscope": The technique can extract information about distributed local photoresistance and field hotspots in thin films, a task previously requiring advanced near-field microscopy.
• Potential Applications: Beyond imaging, the architecture could function as an in-sensor classifier, where unique boundary voltage patterns correspond to recognized images, or be used for real-time spectral and polarization sensing.
• Speed Optimization: While current demonstrations used sequential scanning, future designs could decouple injection and readout contacts to enable parallel readout, significantly improving response times.

In summary, this work demonstrates a paradigm shift in image sensing, replacing complex per-pixel electronics with a simple, scalable, and algorithmically reconstructed resistive network.