Introduction

Mid-wavelength infrared detectors are widely employed in a multitude of fields1, including missile early warning and tracking, night vision imaging2,3, target recognition4,5, environmental monitoring6, astronomical observation7,8, scientific research9,10, as well as medical and biological applications5,11. However, the optical absorption caused by carbon dioxide and water vapor in the atmosphere significantly attenuates the system signal12. Additionally, achieving accurate long-range detection and tracking of weak targets in low-light and complex environments requires devices capable of amplifying signals during the detection process to enhance the signal-to-noise ratio (SNR) at the receiver output. Avalanche photodiodes (APDs), which integrate both the detection and intrinsic gain stages into a single device, greatly improve detection accuracy and resolution, thereby playing a critical role in such weak signal application systems. However, traditional MWIR APDs are typically operated at liquid nitrogen temperatures due to the uncontrollable tunneling current suffered by narrow bandgap devices in response to large multiplication gain13,14,15, the dark current is still the main limiting issue for current MWIR APDs. Moreover, MWIR APD systems are also appealed to operate at higher temperatures to reduce cost and power consumption, making them more suitable for demanding environments and paving the way for extending their versatility in the commercial, scientific and civilian applications16,17. But more importantly, the APDs inherently contribute to an increase in shot noise, \(\left\langle {i}_{\,{\mbox{shot}}\,}^{2}\right\rangle =2q{M}^{2}({I}_{{{{\rm{ph}}}}}+{I}_{{{{\rm{d}}}}})F(M)\Delta f\), by generating an excess noise factor, F(M) = kM + (1−k)(2−1/M), where k is the ratio of electron impact ionization coefficient α to hole β, due to the stochastic nature of the impact ionization process18,19. Therefore, the exploitation of the next-generation MWIR APDs must be equipped with both low dark current, high temperature operation (HOT), and low excess noise factor.

The popular InAs and Hg1−xCdxTe homogeneous p-i-n devices are widely used for MWIR detection and amplification that exhibits a high gain and low excess noise at liquid nitrogen temperature20,21. However, when operating at higher temperatures, the devices reveal poor band-to-band tunneling current for InAs13 and high diffusion dark current due to the low p-type doping concentration (<5 × 1016 cm−3) for HgCdTe22,23.

The Sb-based materials, such as AlInAsSb19, AlGaAsSb24, AlAsSb25, feature a large spin-orbit splitting energy (Δso) to bandgap energy ratio, which can significantly reduce the holes impact ionization coefficients β, representing the most promising candidates for achievable low excess noise factor APDs. They combine the merit of separate absorption, charge, and multiplication (SACM) structure, which maintains a large electric field in a wide bandgap-multiplication region to achieve the high gain, while keeping the electric field in the narrow bandgap-absorption region below the threshold of the tunneling current, to realize a high-performance 1550 nm and 2 μm detection APD devices. The researchers also attempt to extend the Sb-based APD’s operating wavelength toward the mid-infrared. However, APDs still suffer from large dark current when operating at wavelengths above 3 μm, and there is no proven way to address it26,27,28.

In this paper, we present the modified fully-depleted absorption, multiplication (MFDAM) high-temperature-operation HgCdTe MWIR APD structure by employing the bandgap-engineering. The device represents the breakthrough of the dark current density and gain-normalized dark current density (GNDCD) at high gains and high temperature over 160 K. At 160 K, the device exhibits a multiplication gain of 189 with extremely low excess noise F < 1.4, a noise equivalent power (NEP) of less than 7.2 × 10−16 W/Hz1/2 at 3.5 μm and an excellent GNDCD below 2 × 10−6 A/cm2. The imaging results demonstrate that the APD device is capable of capturing signals for remote sensing at distances exceeding 200 km. These findings reveal that the MFDAM APDs are promising for future high-temperature-operating MWIR detection and likely the photon-starved applications.

Results

Device and design

The specific structural details of the heterostructure MFDAM APD is illustrated in Fig. 1a. Processes of the material growth and device fabrication are provided in the Sec. “Methods”. In contrast to the SACM structure commonly employed in III-V group materials, which features a narrow bandgap in the absorption region and a wide bandgap in the multiplication region. The MFDAM device is accomplished by creatively incorporating composition-graded multilayer wide-bandgap Hg1−xCdxTe materials within the absorption region along with composition-graded narrow-bandgap Hg1−xCdxTe materials within the multiplication region, the distribution of Cd composition and cutoff wavelength relative to the device depth is shown in Fig. 1b. Here, the energy bands in which they meet are smooth and continuous to ensure that mid-wavelength photons move into the multiplication region to acquire energy for impact ionization events. Compared with the standard device, this device is unique in that the grading is completely covering the whole structure, which makes sure that the benefits of the different compositions are well appreciated in the different layers. On the one hand, the introduced wide-bandgap gradient material in the absorption region is to generate a built-in quasi-electric field, which is different from the SACM structure but attempts to deplete the layer over a composition grading and serves to mitigate the influence of diffusion current. Additionally, the composition-gradient can effectively drive carriers away from surface states, while the larger bandgap of the short-wavelength material reduces the non-radiative recombination of carriers in the absorption region. Together, these two effects contribute to the overall suppression of the device’s dark current. The multiplication region, composed of narrow-bandgap materials with a graded composition, can provide a high electron impact ionization gain and low excess noise while suppressing tunneling current by carefully designing the thickness. In Hg1−xCdxTe material systems, the reduction of Cd composition makes the bandgap energy notably smaller than Δso at longer wavelengths, resulting in the impact ionization rate of holes being greatly reduced15. Thus, exponential gain and low excess noise are enabled. Besides, only graded composition can further minimize the influence of tunneling current for narrow-bandgap materials. Finally, a little thickness of wide-bandgap materials in the device surface is used to effectively reduce the surface leakage current29. Fig. 1c and d show the energy band diagrams for the device at zero bias and large reverse bias. It can be seen that the electric field is mainly located at the multiplication region as the calculated electric field profile in Fig. 1e, while the absorption region is fully depleted by the built-in electric field. As a result, the MFDAM APD device demonstrates that the mid-wavelength infrared response can be up to 4.2 μm under the 80 K temperature due to the combined efforts of all compositions, as shown in Fig. 1f.

Fig. 1: Modified fully-depleted absorption, multiplication region HOT MWIR APD structure.
figure 1

a Schematic diagram of HgCdTe MWIR avalanche photodiodes with fully depleted absorber layer, the material is composed of multiple layers of composition gradients. b The distribution of Cd composition and the corresponding wavelength at 80 K for proposed MFDAM APD, whose absorption region is a wide bandgap (refers to high composition) altered. The absorbed photon-generated electrons (two in conduction band) will go through the three processes: recombination occurs after absorption (one electron remains), drift to the multiplication region for impact ionization (two electrons in (d)), relaxation in collection region (one electron in (c) and two in (d)). The absorbed two photon-generated holes in the valence band will have no impact ionization happens. The blue and pink circles indicate electrons and holes, respectively. e Modeled electric field profile of the structure showing that the absorber layer is depleted due to the presence of a built-in electric field. This phenomenon arises from the energy differences between the conduction and valence bands caused by the interaction of materials with different bandgaps. The diffusion of carriers leads to band bending, resulting in a redistribution of carriers near the contact surface, which, in turn, forms a space charge region and generates an electric field at the interface. f The normalized spectrum response of the proposed MWIR MFDAM APD device at 80 K.

Full size image

Dark current density and gain

Figure 2(a) displays the measured dark current density of the 100 μm  × 100 μm APD device under 80 K and 160 K. It can be noted that the dark current density is lower than 1 × 10−9 A/cm2 at bias  < −6 V of 80 K due to the presence of wide-bandgap composition gradient and significantly improved over previous conventional homojunction devices15,22. Besides, the band-to-band tunneling current is also effectively weakened, which is obviously observed at the high temperature of 160 K. In order to evaluate the ability of MFDAM MWIR APD device to operate at elevated temperatures, the dark current density of 160 K is quantified to provide a distinct comparison to the theoretical n/p diffusion-limited benchmark30,31. The results are shown in Fig. 2b, it can be clearly identified that the dark current density is nearly an order of magnitude lower than the diffusion limitation at unit gain bias and corresponding to it at large gain M = 5.8. The temperature-dependent dark current density is also provided in Supplementary Fig. S3, the results also exhibit excellent performance even if the operating temperature continues to improve. Besides, the activation energy, which is about half the bandgap, indicates that the proposed APD device is dominated by the generation-recombination (G-R) current at the high temperature. Improvements in minority carrier lifetime to reduce the G-R current will be a benefit to the device’s future operation at higher temperatures.

Fig. 2: Dark Current and multiplication gain measurements.
figure 2

a Measured dark current density versus voltage curves of a 100 μm  × 100 μm device at 80 K and 160 K temperature. b Comparison of dark current density at different gains with the theoretical limit of electron diffusion current at 160 K. It is always less than the diffusion limit with a gain M < 5.8. c The measured multiplication gain along with the reverse bias at the temperature 80 K–160 K for 100 μm  × 100 μm device. The maximum gain is M = 189 at 160 K. d The calculated GNDCD at 80 K and 160 K, the dark current density and gain are from (a) and (c).

Full size image

The multiplication gain is characterized by phase-sensitive technology in “Methods”, the measured gain along with the reverse bias at the temperature 80 K–160 K for 100 μm  × 100 μm device is shown in Fig. 2c. It is discovered that the overall gain decreases as temperature increases, which is attributed to the decrease in the impact ionization coefficient due to the increase in the bandgap as well as the electrons effective mass. The exponential gain is responsible for the pure electrons impact ionization, as explained in previous reports15. A high gain of 189 at -13 V bias is also achieved under the 160 K high temperature. With a bias greater than −13 V, significantly higher gains can be expected, reaching to 1000 at low temperatures of 80 K.

The general evaluation of the APDs is usually examined by the gain-normalized dark current density (GNDCD), which is determined by the measured dark current density and multiplication gain. Fig. 2d shows the GNDCD versus the reverse bias at 80 K and 160 K. The MFDAM APD performed the lowest GNDCD of 4 × 10−10 A/cm2 under −7 V bias with a gain of 11 at 80 K. Moreover, at high temperatures 160 K, the recorded GNDCD is consistently less than 2 × 10−6 A/cm2, even at large gains of 189. Therefore, it can be inferred from the experimental results that the wide-bandgap composition gradient successfully addresses the limitations of the dark current for high-temperature operation, while the narrow-bandgap gradient also enables a large gain at high temperatures.

Excess noise

The measured excess noise factor (F) of the MFDAM structure is revealed as a function of gain, as shown in Fig. 3a. The design exhibits extremely low excess noise below the local field model limitation of 232, which can be explained and developed by the latest non-local field theory33. The excess noise factor is of 1.39 at the gain of 160 for 160 K operating. In the lower gain range, the excess noise increases as the fact that the average impact ionization length remains greater than the dead space, and the increase of uncertainty in the impact ionization process15,33. The low excess noise Sb-based APDs24,26,34,35 and previous HgCdTe APD29 are employed to compare with the MFDAM structure, results are concluded in Fig. 3(b). The MFDAM APD is proved that can carry out higher gain at elevated temperature as well as the lower excess noise. This is mainly due to the fact that narrow-bandgap gradient can effectively hold down the tunneling current to ensure that the device can be operated at a higher bias voltage.

Fig. 3: Excess noise measurements.
figure 3

a The excess noise versus multiplication gain of the proposed 100 μm  × 100 μm APD device at 80 K, 160 K, and 200 K. b The comparison of different low-excess-noise APDs at different wavelengths and working temperatures with this work. The color of the solid circles represents the working temperature shown in the temperature bar, and the label around the circles indicates the cutoff wavelength.

Full size image

Quantum efficiency

Spectral response measurements were used to determine the external quantum efficiency (EQE) of the extended MFDAM APDs. Fig. 4a displays the spectral EQE and responsivity obtained of an MFDAM APD under -0.2 V voltages at 80 K and 160 K temperatures. The device exhibits a cutoff wavelength of  ~4.1 μm at 160 K, and the EQE at 3 μm is about 40%, the maximum EQE is of 45% at 2.67 μm wavelength. It should be noted that the lateral collection length that comes from the neighboring pixel is also considered as depicted in “Methods”. The comparatively low quantum efficiency can be attributed to the reduced thickness of the absorbing composition, as illustrated in Fig. 1b, Incident photons do not travel a sufficient distance to be well absorbed, particularly in the mid-infrared region where the photons diffusion length is relatively long. Besides, in extending the short-wavelength range, the recombination results in the absorbed carriers being unable to effectively contribute to the photocurrent. Future improvements in quantum efficiency may focus on several approaches, including the incorporation of anti-reflection (AR) coating26 or photon-trapping35 as potential solutions. The fundamental objective is to enhance the thickness of the absorbing composition while minimizing any adverse impact on the dark current.

Fig. 4: Quantum efficiency and noise equivalent power measurements.
figure 4

a Measured unity-gain external quantum efficiency and responsivity of a 100 μm  × 100 μm APD device at 80 K and 160 K. b The noise equivalent power versus wavelength under 292 gain for 80 K and 189 gain for 160 K. The noise is based on the measured dark current under the −13 V bias.

Full size image

The noise equivalent power (NEP) is a widely used figure of merit for APD-based receivers, representing the minimum detectable light signal. Consequently, a lower NEP indicates superior performance. In this MFDAM APD device, the NEP is calculated by the measured dark current, gain as well as the responsivity, the results under −13 V are all shown in Fig. 4b. For 160 K and 189 gain, the NEP is below  < 7.2 × 10−16 W/\(\sqrt{Hz}\) of 3.5 μm wavelength, the value is slightly increase for wavelength low than 3.5 μm as the responsivity decrease, but the obtained results are still less than the 2 × 10−15 W/\(\sqrt{Hz}\). These figures all point to the fact that the MFDAM APD device is an excellent performance at high temperature 160 K.

Weak signal imaging

Next, to extend the applications and prove the performance more balanced, a small arrays that produced with MFDAM MWIR APDs are used to image targets in complex contexts at remote distances, since the high performance APD device can achieve improved signal-noise-ratio (SNR) due to the internal multiplication gain compared with the normal MWIR device. In this work, we demonstrate the devices for imaging at gains below 10, where they have a low dark current density and NEP as well as high uniformity. Figure 5 presents the imaging results at varying distances while a civil aviation aircraft passes overhead. It can be observed that the target is located at 80 km, where the SNR is 93 dB, a clear pattern is obtained including the tail flame. As the distance increases, the signal strength diminishes, leading to a greater prominence of noise contributions from the sensor and the surrounding environment. However, the advanced design of our MWIR APD devices makes it possible to increase the signal value without significantly introducing heavy dark currents and noise. Thus, the tail flame can be visualized at target distances greater than 200 km, even if the environment is severely attenuating the signal and the field of view is reducing.

Fig. 5: Imaging of targets at long distance.
figure 5

The proposed MFDAM APD devices are utilized for imaging weak targets at varying long distances of (a). 80 km, (b). 100 km, (c). 160 km and (d) 200 km.

Full size image

Discussion

Table 1 presents a comparative analysis of our work alongside other emerging III-V MWIR SACM APDs, including an AlInAsSb-AlInAsSb26,36,37, InAs/GaSb-AlAsSb27, InAs-AlAsSb28, InGaAs/GaAsSb-AlGaAsSb34, respectively, as well as the traditional InAs38, HgCdTe30 APDs. It is evident that our device stands out for its performance in all metrics. Especially, the longer cutoff wavelength of 4.1 μm in our structure versus the AlInAsSb APD (<3.5 μm) means that more infrared information can be obtained at comparable temperatures to satisfy the extensive applications domains. The lower gain normalized dark current density ensures that the APD operates at high gains without cooling down the temperature. There is also an order of magnitude advantage over HgCd0.33Te (0.039 mA/cm2) in the same condition. In general, this device departs from the traditional SACM structure, which employs the short-wavelength material in the multiplication region to suppress the tunneling current. Additionally, the MFDAM APD, with a fully depleted short-wavelength absorption region, can ensure that the device’s dark current is primarily governed by the generation-recombination process. Besides, the short-wavelength region exhibits a lower intrinsic carrier concentration ni and a longer carrier lifetime τSR, which can diminish the generation-recombination current (\({J}_{gr}\propto {n}_{i}\times 1/{\tau }_{SR},vs,{J}_{diff}\propto {n}_{i}^{2}\times (1/{\tau }_{A}+1/{\tau }_{SR})\)) that is significantly superior to the diffusion current and thus prove transformative in achieving enhanced performance for mid-wavelength infrared devices.

Table 1 Summary of the Reported MWIR APDs for HOT
Full size table

Conclusion

In conclusion, we have demonstrated MFDAM APDs with composition-graded HgCdTe materials for a large gain, low excess noise, low GNDCD,  >4 μm cutoff at elevated temperature of 160 K, which was previously deemed not possible for MWIR APDs. The whole performances are improved by optimizing the distribution of composition and multiplication region parameters. At 160 K, our device exhibits the GNDCD preserves consistently below 2 × 10−6 A/cm2 for gain values less than 189, while the excess noise holds below 1.4 and the noise equivalent power is  <7.2 × 10−16 W/Hz1/2 of 3.5 μm. This device achieves GNDCD and excess noise better than previously best MWIR APDs, and these results are promising for high-temperature-operation MWIR APD detection applications, in the meantime, defending the absolute dominance of HgCdTe-based APDs. For the future, improving quantum efficiency and optimizing minority carrier lifetimes are at the heart of the efforts.

Methods

Material growth and device fabrication

The n-type low doping concentration (<5 × 1014cm−3) Hg1−xCdxTe thin film was grown on an infrared transparent (111) CdZnTe substrate using vertically liquid-phase epitaxy (LPE) technology. As the Cd composition difference between the CdZnTe and Hg1−xCdxTe interface on the back of the material, the high-temperature and long-time inter-diffusion annealing was employed to deliver the grading to cover the whole structure. The CdTe/ Hg1−xCdxTe interface on the surface of the material was also treated, as shown in Fig. 1, the composition distribution of the final material. The XRD patterns are provided in the Supplementary Fig. S2 to analyse the matrial quality. The p+ absorption layer of the device was doped by Hg-vacancy with 1~2  × 1016 cm−3. The n+ region was obtained by B+ ions implantation. Through a certain period of rapid annealing treatment, the n− multiplication region of low doping (<5  × 1014 cm−3) could be formed. The device was then covered by a ZnS/CdTe double passivation layer to suppress the surface leakage current. The mesa is corroded by the 0.3% HBr. Finally, Cr/Au (50/150 nm) was deposited for n-contacts and p-contacts using an e-beam evaporator. The final device is pictured by the SEM in Supplementary Fig. S1.

Current-voltage characterization

Current-voltage (I-V) characteristics were measured with a Keithley 6430 source meter under blackout condition at different temperature for various device size, as shown in Supplementary Fig. S3.

Multiplication gain

The net photocurrent was measured with a blackbody source (T = 700 °C), a Fast Fourier Transform (FFT) analyzer, and a low-noise preamplifier. A mechanical chopper modulated the blackbody light at 130 Hz to remove DC dark current, and a Keithley 2901 source meter unit was used to apply the bias. The multiplication gain M(V) was calculated as the ratio of the net photocurrent (Inet) at bias V to that measured at the unity gain point of −0.5 V. It was noted that the gain calculation has considered the slight increase of injected photocurrent with reverse bias due to the widening of the depletion width given the long diffusion length of HgCdTe materials.

$$M(V)=\frac{{I}_{{{{\rm{net}}}}}(V)}{{I}_{{{{\rm{net}}}}}(-0.5)}$$
(1)

Excess noise

The proposed MFDAM MWIR APD was biased at different temperature with a Keithley 2400 source meter. A 1550 nm continuous-wave laser was used to illuminate the APDs. The alternating current component of the output current was measured with an Agilent 8970 B noise figure analyser with a 50 Ω load impedance through a standard radio frequency bias tee. Careful system calibration was carried out via a calibrated Agilent 346A noise source to remove background noise. Devices were measured over a 4 MHz bandwidth centered at 20 MHz. Supplementary Fig. S4 describes the linearity of photocurrent versus normalized noise power for different light intensities at unit gain. The slope Aslope of the linear fit is used to calculate the excess noise factor for different gains at constant optical power.

$${S}_{unit}=2qB\cdot {I}_{unit}={A}_{slope}\cdot {I}_{unit}$$
(2)
$${S}_{M}=2qB\cdot {I}_{unit}{M}^{2}F={A}_{slope}\cdot {I}_{unit}\cdot {M}^{2}F$$
(3)

Quantum efficiency

The response spectrum and quantum efficiency of the backside-illuminated device were measured by a Fourier transform infrared spectrometer (FTIR) and calibrated by a 700 °C blackbody source. The calibration of the responsivity includes the effect of the lateral collection length, which correlated with the crosstalk signal of neighboring pixels in the array and resulted in the optical collection radius (Lc) being larger than the actual device radius (Lr). Here, the lateral optical length Lc could be determined by measuring the photocurrent Iph of the diodes with different device radius Lr using the following equation, where the Iph−1D is the photocurrent collected by a diode with no lateral collection. The fitting result based on the different device radius is drawn in Supplementary Fig. S5.

$${I}_{ph}={I}_{ph-1D}\cdot {\left(1+\frac{{L}_{c}}{{L}_{r}}\right)}^{2}$$
(4)

Numerical simulation

The theoretical model was established with a commercial software package Silvaco TCAD. The device modeling was developed on the basis of the distribution of composition, which was carried out by the secondary ion mass spectroscopy (SIMS). The energy band structure and electric-field intensity profile were all obtained.