Delivering multi-band detection with heterogeneous arrays
BY LAXMY MENON, HONGJUN YANG AND WEIDONG ZHOU FROM THE UNIVERSITY OF TEXAS AT ARLINGTON, MATTIAS HAMMAR FROM KTH, SWEDEN, AND ZHENQIANG MA FROM THE UNIVERSITY OF WISCONSIN-MADISON
It's rare that one material system can tick all the boxes. So it can make sense to unite two or three. That's the case with triple-junction solar cells, and it is an option for next-generation ICs and photonic integrated circuits.
Uniting different materials is also a winning formula for imaging systems. Here, integrating materials with different bandgaps enables detection in several wavelength ranges. This holds the key to the construction of multi-band, multi-spectral imagers that can be used for remote sensing, industrial surveillance systems, bio photonics, automotive cameras, fluorescent imaging, spectrometers on a chip, LADAR, and free-space optical communications.
Recently, the greatest interest in this sector has been associated with the simultaneous acquisition of visible and infrared light. There are several ways to do this, and the three most popular options today are using colour filter arrays, external mechanical filters, and exploiting a material's depth-dependent absorption.
With the first of these three, the colour filter array, different colours are identified by placing an array of spectral filters over an imager sensor. The traditional approach has been to have 50 percent of the area covered by a green filter, and the remainder split equally between red and blue.
Recently, however, there has been a move to combine red, green, blue and panchromatic, the latter of which is sensitive to all. But there are downsides of this approach that stem from the high sensitivity of panchromatic pixels to visible wavelengths, and the high sensitivity of the red, green and blue cells to near-infrared wavelengths.
Figure 1. Crystalline nano-membrane-based stacked multi-colour, multi-band photodetector arrays: (a) Three-junction silicon-on-insulator wafer, optimized based on the penetration depth of blue, green and red wavelengths in silicon, and an InGaAs single junction wafer, before transfer; (b) silicon pixels transferred onto an InP substrate with precise alignment with an InGaAs pixel; (c) one pixel of multi-colour, multi-band membrane imager after device completion.
If the second option is applied "“ the use of mechanical filters "“ then the most common approaches are to capture the same scene through two separate cameras simultaneously; or to use an IR blocking filter, or a visible blocking filter, to simultaneously take two images of the same scene. Unfortunately, both approaches require substantial post processing, such as image fusion. In addition, hardware is expensive, bulky and slow, whether filters are replaced mechanically or two cameras used. What's more, greater processing power is needed to analyse and compare two separate images.
For the third option, which involves exploiting the depth-dependent absorption property of a material, the leading technology is based on the vertically integrated photodiode structure. InGaAs and silicon is a common combination, with the quantum efficiency of InGaAs increased in the visible band through removal or thinning of the InP substrate.
Transfer printingOur team at the University of Texas at Arlington is pursuing this type of approach. We are pioneering the use of a transfer printing technique to construct a multi-band detector based on silicon and InGaAs.
It is possible to build a multi-colour detector with just silicon, because photons with different energies have different absorption depths. The way to do this is to stack several p-n junctions on top of one another and exploit an absorption depth that varies by several orders of magnitude over the visible range. Merits of this approach include avoiding the inevitable transmission losses and colour aliasing effects, and having a resolution that is limited by colour-filter-based sensor arrays. There is a limit to what silicon can do, however. Its bandgap prevents detection at wavelengths beyond 1.1 µm, so we unite it with InGaAs, which is capable
of detecting near IR wavelengths up to 1.68 µm.
Two widely used approaches for uniting different semiconductor materials are epitaxial growth and wafer bonding. Both have weaknesses: epitaxial growth, for example, cannot unite all materials and requires high temperatures, while wafer bonding is incompatible with some forms of semiconductor (see table 1 for a more detailed account of the limitations of both techniques).
Table 1. Comparison between polydimethylsiloxane (PDMS) printing, wafer bonding and epitaxial growth.
We overcome all these shortcomings with a transfer printing process that has been widely used to create heterogeneously integrated devices. The process involves a polydimethylsiloxane (PDMS) stamp that acts as the intermediate medium, transferring a membrane layer from the parent substrate to its new home. The great strength of this approach is that the highest performance material for a specific application can be transferred or integrated into unusual and challenging environments. Devices made by various groups include a membrane-reflector VCSEL, solar cells and an n-channel MOS inverter.
In our case, we form our heterogeneously integrated multi-band photodetector array by picking up silicon chiplets from a silicon-on-insulator platform with a PDMS stamp and transfer-printing them on to InGaAs photodetectors (see Figure 1).
The silicon chiplets have a vertical n-p-n-p structure, and are capable of detecting blue, green and red, thanks to the wavelength dependent absorption property of silicon. Near infrared light is captured by the InGaAs detector, which has ap-i-n structure with a 2 µm intrinsic region. The transfer process must provide precise alignment, in order to ensure an alternately placed silicon
and InGaAs photodetector array (see Figure 2).
With our approach, there is no need for mosaics, or for interpolation. What's more, by capturing visible and infrared radiation in a single shot, with the same resolution, we avoid complicated processing steps associated with the deposition of infrared blocking filters at the pixel level.
Proven successPerforming current-voltage measurements on our silicon and InGaAs devices verifies their good diode characteristics. We have also subjected these devices to different illumination wavelengths. The photoresponse of the devices show significant variation with incident intensity, indicating good responsivity (see Figure 4). This is encouraging, because responsivity is directly related to device efficiency.
If a multi-band detector is to offer good resolution during imaging, all its devices must deliver a uniform performance. When this occurs, colour images can be constructed by measuring the spectral responsivities for each junction and then performing post-imaging data processing.
Our development of multi-band detectors is in its infancy, with efforts beginning with a 4 by 4 array. With just 16 pixels, imaging an object is not possible.
So instead we have evaluated the performance of our detector by trying to recreate a column of pixels illuminated by a narrow slit, by imaging. In this process, we illuminate only one column of pixels through a narrow slit, while all other pixels remain dark. The photoresponse of all the pixels (all junctions) is measured in this condition, and by using the data obtained in post processing techniques, we can image the illuminated column of pixels. On the array, the pixels that are not illuminated by the light through the slit remain dark.
Figure 3. Measured current-voltage characteristics of blue, green and red junctions of silicon under illumination with (a) 405 nm (b) 532 nm (c) 632 nm, revealing responsivities of 0.09 A/W, 0.1 A/W, and 0.15 A/W, respectively. Measured responsivities in an InGaAs photodiode at 980 nm and 1550 nm are 0.5 A/W and 0.8 A/W.
This experiment was successful, with the illuminated column "“ column 2 for the blue junction, and column 3 for green, red and infrared junctions "“ easily distinguished from the non-illuminated columns (see Figure 4). The non-uniformity in colour is due to a variation in the intensity of the light hitting specific junctions. This is promising, as it lays the foundation for imaging an object, or scenery during day or night.
We have shown that by utilizing the inherent properties of silicon and InGaAs, and uniting them with an established PDMS transfer printing method that does not require any expensive equipment, it is possible to build a lightweight, cheap, heterogeneous detector array. Other semiconductor materials could be used in this process, enabling the production of multi-band detectors covering other spectral ranges, and providing spectrally resolved sensing systems for other applications. We are already working on creating a GaN and silicon heterogeneously integrated light source/photodetector for bio-medical applications.
Figure 4. Reproduced images of the illuminated column in the 4à—4 array formed by biasing the (a) blue junction; (b) green junction (c) red junction and the (d) infrared junction. The horizontal and vertical numbers in the figure indicate the row and column number of the array. Each square in the figures represents the difference between the measured photo current and dark current for that specific junction of the pixel. The scale bar on the right side of each figure shows the range of values in the square. Non-uniformity in colour is due to the variation of this diference between the various pixels, for that specific junction.
To enable our technology to serve even more applications, we are trying to build a detector on a flexible substrate, such as kapton or a shape-memory polymer. Additional goals are to increase the size of our arrays and their fill factor.
