1. Introduction

Digital imaging has until recently been dominated by Si based arrays, which are only able to detect visible or near infrared light due to the bandgap of Si. As such imaging in the short and mid wave infrared (SWIR, MWIR) regions of the spectrum has not been as intensively exploited. Imaging in the SWIR offers potential benefits such as imaging using reflected star light [1] which peaks in intensity at a wavelength of 1.7 μm, or active imaging using a pulse from an eye-safe 1.55 μm laser to illuminate a dark scene and image the reflection [2]. Imaging in the MWIR can also be used for radiation thermometry applications, such as to monitor the manufacturing temperature of plastics. In addition there are a number of imaging applications which require a detection wavelength around 2 µm including; 3D LIDAR detection for CO₂ in both ground-based and airborne applications [3–5], imaging the diffusion of Glucose [6,7] and ‘cold’ radiation thermometry to identify regional hot spots [8].

Due to its bandgap InAs can respond to wavelengths from the visible up to the MWIR, which makes it highly attractive for use in this class of imaging equipment. Furthermore over recent years InAs has been demonstrated to operate as an excellent avalanche photodiode (APD) over this spectral range with single carrier multiplication and low excess noise [9,10]. Alternative semiconductor materials that could be utilized for imaging applications over this wavelength range include Cadmium Mercury Telluride (CMT) [11,12], the band gap of which can be tuned via its composition from 1 to 16 µm, with high gain and low noise. Additionally CMT APDs have been successfully demonstrated in Focal Plane Arrays [12]. However CMT is an expensive and difficult semiconductor to grow and fabricate, limiting both the price and quantity of devices that are available. CMT APDS have previously demonstrated high gains at longer wavelengths (a gain of 100 has been reported with an applied bias of 5 V when operating at 9 µm) [13]. However when the composition is changed to lower the cut-off wavelength the gain also reduces (when operating at 2.2 µm a gain of less than 2 is achieved at 5 V) [13]. GaAsSb/InGaAs Type-II superlattices have also been successfully demonstrated as photodiodes operating in the 2 – 3 µm range. Furthermore they can be operated in an avalanche mode by forming a separate-absorption-and-multiplication (SAM) APD, where the superlattice forms the absorber and InP [14] or InAlAs is used for the multiplication [15]. APDs utilizing the bulk semiconductor InGaAsSb have also been realized over this wavelength range [16], however the growth and fabrication of this material is still at a very early stage with devices dominated by large surface currents. Consequently no APD arrays have been reported from the Type II and InGaAsSb devices to date.

Most of the recent work in InAs has been focused on improving fabrication and growth techniques for mesa diodes, as evidenced by the most recent publications being on studies of leakage currents [17,18], with the best results being obtained using wet chemical etchants and photoresist based passivation. However it is not clear at present whether these fabrication techniques are robust and reliable enough to allow the realization of a uniform array which could be deployed for imaging applications.

In this paper we report on the first 128 pixel linear array of InAs APDs which have been fabricated using standard mask lithography and wet chemical etching techniques. We have investigated both the electrical and optical uniformity of the array as well determining the degree of electrical isolation between the pixels.

2. Experimental details

Due to the disparate ionization coefficients in InAs only electrons initiate impact ionization at practical electric fields below ~70 kV /cm, before the onset of band to band tunneling currents [9,10]. Therefore to ensure electron injection into the multiplication region the incident signal should be fully absorbed in the p-doped layer in a p-i-n based APD design. However for use in imaging applications using 2D arrays, the reverse structure, a n-i-p design, would be grown to allow the pixels to be bonded via a flip-chip technique for integration with the read out electronics before the substrate is removed to allow illumination on the p-layer. For this reason the structure investigated in this work is a n-i-p configuration with an n⁺ top layer. The structure was grown via solid source molecular beam epitaxy (MBE) on a 2-inch p + InAs substrate at 510 °Celsius with a growth rate of 0.6 ML/s. The high growth temperature used here was chosen to increase the growth rate so that impurity incorporation can be kept low [17]. In and As₂ fluxes were supplied from a dual filament effusion cell and a valved cracker respectively. An As₂/In beam equivalent pressure ratio of 10 was used for the whole structure. The substrate was rotated at a speed of 30 rpm to minimize the effect of heating and flux non-uniformity. The structure consists of a 1.5 µm p⁺ doped (1x10¹⁸ cm⁻³ of Be) layer followed by a 4 µm undoped layer and then a 2 µm n⁺ doped (1x10¹⁸ cm⁻³ of Si).

The pixel array was defined via standard contact mask lithography (to form square pixels with dimensions of 50 µm square) and the pixels were etched using our preferred chemical etching procedures of a 1:1:1 (phosphoric acid: hydrogen peroxide: de-ionized water) etch, followed by a finishing etch of 1:8:80 (sulfuric acid: hydrogen peroxide: de-ionized water) [9]. The diodes were then passivated with SU8, which has previously been found to suppress surface leakage currents in large area mesas down to 77 K [6]. The SU8 layer also formed a dielectric layer upon which the bond pads were deposited. The bond pads and top contacts were deposited simultaneously in the same step and were formed from Ti/Au with thicknesses of 20/200 nm. A common p contact was formed using 20/200 nm of Ti/Au. An SEM image of a section of the array is shown in Fig. 1, each pixel had a fill factor of 67%.

 

Fig. 1 (a) SEM image of a section of the fabricated array, (b) magnified image of pixel and bond pad and an optical image of the array

Download Full Size | PDF

Measurements were performed on a Janis ST-500 probe station utilizing two DC probe arms for electrical contact and a multimode fiber to provide optical excitation. Current-Voltage (IV) measurements were performed by connecting the probe arms to a Keithley Source Measure Unit (SMU). For measurements of the photocurrent and avalanche gain a nanoplus DFB laser with an output wavelength of 2.04 µm was coupled into the multimode fiber giving an incident power of 1.25 mW onto the sample. The avalanche gain was determined via the extrapolation of the primary photocurrent at low biases [19]. As each pixel of the array had to be individually probed and aligned to obtain measurement data, the majority of the avalanche gain measurements were taken under DC methods, where the combined dark and photocurrent was measured with the SMU and then the dark current was subtracted from this total current to obtain the photocurrent. To ensure accuracy in our measurements these results were confirmed for selected pixels via the use a lock in amplifier and phase sensitive techniques, which measures the photocurrent independent from the influence of dark current.

3. Results and discussion

To assess the uniformity of the array we measured the IV characteristics of each individual diode in the array with the results shown in Fig. 2. Figure 2(a) shows that the full IV measurements for a selected pixel (several pixels were compared but only one shown for clarity) at three temperatures. At all temperatures the IV characteristics are very similar indicating highly uniform wafer growth and fabrication processes. The IV results for a large area diode (circular mesa with a radius of 100 µm) which has been fabricated from the same wafer is also included for comparison. The current densities, for both the pixels from the array and the large area diode, are in excellent agreement at room temperature indicating that the dark currents in these diodes are dominated by bulk mechanisms. However as the temperature decreases an increasing discrepancy between the array pixels and large area diode is observed indicating the presence of surface leakage at lower temperatures.

 

Fig. 2 (a) IV characteristics of a selected pixel (symbols) and a reference large area mesa (dashed lines) at temperatures of 295, 200 and 77 K, the 295 K photocurrent is also plotted (solid line), (b) dark current as a function of pixel number at fixed voltages.

Download Full Size | PDF

To further investigate the uniformity of the array Fig. 2(b) shows the measured dark current density as a function of pixel number at a fixed bias at room temperature. From Fig. 2(b) it can be seen that the array has a good uniformity with the measured current, with only approximately a 5% variation in the majority of pixels. Larger variations are observed on either end of the array (i.e. 10 pixels on each end). This increased variation in current is believed to be originating from the fabrication process due to the relatively small size of the sample. The array was fabricated on a piece of wafer with dimensions approximately 2 mm longer than the total length of the array, resulting in only a 1 mm gap between the last pixel and the edge of the wafer as shown in Fig. 1(b). As such the end pixels may be degraded due to edge bead effects, cleave planes and physical sample handling.

The room temperature mean dark current of 0.5 Acm⁻² at 0 V bias is higher than that reported for other materials operating in the SWIR region including CMT operated at 2.2 μm wavelength (5x10⁻⁴ Acm⁻² [13]), and extended InGaAs (i.e. Hamamatsu G-12183-10K which has a 0 V dark current of ~1x10⁻⁴ Acm⁻²). While the dark current at punch through for a type two superlattice based SAM-APD is ~5x10⁻³ Acm⁻² [15]. Although the dark current in this work is relatively high, a much better InAs dark current density of ~3x10⁻² Acm⁻² from a single diode was reported [20], indicating that further improvement should be possible. It is important to note that all of the above materials have a shorter cutoff wavelength than that reported here, so InAs having a larger current density is neither a major surprise nor an indication of competitiveness at room temperature. We anticipated that InAs APDs could achieve high performance at temperatures close to 200 K. It is therefore necessary to cool the device to lower the dark currents and allow low levels of the photocurrent to be detected. Hence we have evaluated the uniformity of the dark current of the array at 200 and 77 K with the results shown in Figs. 3(a) and 3(b) respectively. Due to the increasing presence of surface currents as observed in Fig. 2(a) at these temperatures, these results are displayed in terms of current, as the true bulk current density cannot be accurately determined.

 

Fig. 3 Current as a function of pixel number for selected voltages at 200 (a) and 77 K (b).

Download Full Size | PDF

From Fig. 3 it can be seen that at both temperatures the array still has a high uniformity with an increased variation in current, again only occurring at the edges.

Having established that we have fabricated a uniform array we evaluated its performance as an array of APDs. For these measurements we have measured the avalanche gain of each pixel in the array at a wavelength of 2.04 µm at a temperature of 200 K by coupling the laser emission into a fiber and focusing it onto each pixel. Since the wafer structure is an n-i-p configuration but we require electron injection to initiate impact ionization in InAs [9] the laser emission from the fiber (with core diameter 62.5 µm) was shone onto the p + layer, at the bottom floor, of each pixel to achieve predominantly pure electron injection. Due to practical limitations concerning the sample and experimental set up, this method of performing the analysis of the array may have some mixed injection which could lower the gain at a given bias. The photocurrent was measured under a DC technique where the total current (dark and photocurrent) was measured via an SMU.

The measured avalanche gain at 200 K is shown in Fig. 4, where Fig. 4(a) shows the gain curves for selected pixels while Fig. 4(b) shows the gain at a fixed bias as a function of pixel number. The gain curves observed are typical for InAs due to impact ionization being initiated only by electrons [9] where the onset of gain is observed at very low voltages and the gain rises exponentially with voltage. The gain curves for all of the pixels show a high level of agreement (as illustrated with the selected pixels in Fig. 4(a) illustrating that the sensitivity of the gain is the same for all pixels. At zero bias, the measured photocurrent results in a calculated responsivity of 0.6 A/W.

 

Fig. 4 (a) DC gain characteristics of selected pixels (symbols) and gain determined using phase-sensitive detection (dashed lines), (b) gain as a function of pixel number at fixed voltages all performed at 200 K.

Download Full Size | PDF

At a voltage of 10 V an avalanche gain of approximately 8.5 is measured, from Fig. 3(a) the associated dark current with this gain is 15 Acm⁻². Measurements were limited to this voltage due to these high dark currents in the sample making it difficult to accurately extract the photocurrent using a simple subtraction of dark current from the total current. The same level of gain has been obtained in SWIR CMT under room temperature operation with a lower associated dark current (7x10⁻⁴ Acm⁻²) [13]. Similarly GaAsSb/InGaAs superlattice based SAM-APDs produced dark current density of 5x10⁻² Acm⁻² [15] at a similar gain at room temperature. As noted previously the InAs diodes reported here do not exhibit the lowest reported dark currents and also suffer from surface leakage currents. We believe lower dark current densities are achievable as values of ~2x10⁻³ and 2.5x10⁻¹ Acm⁻² at 200 and 300 K respectively, have been achieved at the same gain [18]. Furthermore the InAs pixels reported here provide this level of gain at a much lower voltage than either CMT [13] or superlattice based SAM-APDS [15], which should offer easier integration with readout electronics. From Fig. 4b it can be seen that excluding the pixels on the edge of the array there is a high uniformity in the measured gain across the array. This is further supported by the histograms shown in Fig. 5 which shows that at each voltage the individually measured gains are centered closely around the mean value. With 89%, 91%, 88% and 92% of the pixels having a deviation of less than 10% of the mean gain at voltages of 2.5, 5, 7.5 and 10 V respectively. The pixels that fall outside of this range correspond to those on the edges (as observed in Fig. 4(b) and may be due to errors caused by high leakage current.

 

Fig. 5 Histogram distributions of avalanche gain at selected voltages. Measurements were performed at 200 K.

Download Full Size | PDF

Using a DC measurement technique for the determination of the avalanche gain can introduce errors when the ratio of photocurrent to dark current is small. From Fig. 2(a) it can be seen that at room temperature the ratio of photocurrent to dark current is approximately 10 at 0 V. However the ratio decreases slightly as the voltage is increased. As the dark current decreases with decreasing temperature while the photocurrent remains approximately constant this ratio will increase at lower temperatures. To verify that errors have not been introduced into our measurements from this technique we have re-measured the gain of the same selected pixels as shown in Fig. 4(a) using a phase-sensitive detection approach (lines in Fig. 4(a). Due to the nature of our experimental set up which requires each individual pixel to be manually probed and aligned it was not feasible to test the whole array under these conditions. However the selected pixels show an excellent agreement between the two methods, illustrating that our DC method has not lead to significant errors as the photocurrent to dark current ratio is relatively large (for example a ratio of 31 is obtained at 0.5 V at 200 K). Our measurements confirm that under DC method the gain appears to be higher for the pixels near the edges, but phase sensitive measurements show a lower gain which is consistent with that obtained from the center of the array. The reason for these discrepancies is believed to be the increased dark current at the edge which reduced the signal to noise leading to measurement errors, when DC method is used.

As mentioned previously, to achieve pure electron injection into the multiplication region we have focused the incident laser radiation onto the p⁺ layers of the pixels. However this may introduce additional alignment errors when compared to a simpler top excitation in a p-i-n geometry. To determine the influence of optical spot position we have performed a separate measurement using free space optics (with the same laser) at room temperature. A computer controlled micrometer was utilized to move the device position, relative to a fixed laser spot of 10 μm. The initial position is taken as the edge of the pixel as viewed from directly above; the device is then moved linearly away from this point. The measured gain as a function of excitation position, using phase sensitive detection is shown in Fig. 6(a). It can be seen that significant error in gain was measured when the optical spot is more than 15 μm away from the pixel. Inferring from this, we believe that our fiber alignment during gain measurements in Fig. 4 is within 15 µm from the pixel since highly uniform gain was measured.

 

Fig. 6 (a) Room temperature avalanche gain as a function of excitation position (b) Measured photocurrent when light is focused onto a probed pixel (circles) and when focused onto an adjacent pixel (triangles).

Download Full Size | PDF

To evaluate how well isolated the individual pixels are, we have measured the photocurrent from a pixel while focusing the incident radiation on an adjacent pixel. This measurement was again performed at room temperature using the same laser but with free space optics to ensure an optical spot smaller than the pixel. The measured photocurrent responses when the light was focused onto the measured pixel and an adjacent pixel are shown in Fig. 6(b). An approximately voltage independent photocurrent of less than 10 nA was recorded which is over five orders of magnitude lower than the directly measured photocurrent is observed. This indicates that the individual pixels have a good degree of optical isolation and that the small measured photocurrent is thought to be most likely due to stray reflected light.

4. Conclusion

In summary we have fabricated and tested a linear array of InAs APDs with 128 pixels. We have observed good uniformity in the dark currents at all temperatures and a highly uniform gain distribution at 200 K under excitation at 2 µm. These results suggest that InAs is a highly promising material to from infra-red imaging arrays in the future.