MoS₂ based 2D material photodetector array with high pixel density

3.1 Device characterization

The devices are tested over the visible spectrum from 500 to 900 nm at varying optical power levels to see the photo response with the corresponding conditions. Results for these tests show a main peak at 670 nm which increases in relative strength with decreasing optical power level, up to a maximum responsivity of 0.8 A/W (Figure 2a). This suggests that the device is saturating with higher optical power levels, further evidenced by the response curves shown on average decreasing in responsivity with the increasing optical power levels. The primary peak seen in the response corresponds well with main excitonic peak in MoS₂ which has been shown to exist in the region of 1.8–1.9 eV. In addition to the main peak seen at 670 nm a secondary peak can be seen around 600 nm corresponding to the secondary excitonic peak around 2.0 eV. Finally, a third peak at 860 nm is caused by indirect band gap absorption seen in multilayered materials. Another pertinent result from this testing is the decrease seen in the main peak with increasing optical power, suggesting that the saturation of the devices is significant in the direct band transition, however the secondary excitonic peak remains relatively constant throughout shifting optical power levels. Additionally, the indirect band transition sees a significant initial decrease from increasing optical power but less of a decrease at further increasing power levels (Figure 2b). This decrease however does not affect the total function of the device, with the main response shifting to slightly shorter wavelengths in visible spectrum with increased optical power. Furthermore, the red-light regime can still be detected albeit with a lower response than the red–yellow border regime, and the near infrared around the indirect peak remains relatively stable.

Figure 2:

Device and application performance. (a) Responsivity for the optical excitation range of 500 nm–900 nm, for three different laser power levels. Peak photo response is seen growing at the main peak, 670 nm, with decreasing optical power. Further two additional peaks can be seen 610 nm and 860 nm, corresponding to additional photo response peaks seen in multi-layered MoS₂. Inset shows the incident optical power for all three power levels, ranging from 50 nW to 580 nW. (b) Responsivity over changing optical power at 600, 670, 680, and 860 nm. (c) Device responsivity over bending radius, slight increases can be seen with shallow bending radii before decreasing to a maximum bending radius of 1.25 cm. The maximum bending radius was used for cyclability testing between flat and bent positions. Inset shows results over 100 bending cycles, showing less than 10 % variation in responsivity after the first bending cycle of the as fabricated devices. (d) Device responsivity as a function of annealing conditions of the array on both SiO₂ and PI substrates. Annealing was done in an inert environment using a hotplate where arrays were allowed to heat up and cool down while on the plate. n = 16.

Exploring further the device characteristics with changing optical power we observe that as the optical power is even further increased, to microwatt levels, that a roll off is occurring in the device saturation in the main peak (Figure 2b). Additionally, it is shown that the device saturation characteristics are occurring at an earlier power level in the indirect gap transition, where roll off is beginning to be seen in the 10’s–100’s of nanowatt range (Figure 2b). Finally, in the second excitonic peak a relatively consistent response is seen suggesting there’s no significant saturation occurring, although it is possible that the saturation still occurs outside of the tested optical power ranges. This is supported by the fact that the optical power range at the 600 nm wavelength test is slightly lower than that of 670 and 680 nm. Further the upper limit of the 600 nm power range appears to begin dropping in responsivity but needs a larger test span for conclusive evidence.

Testing is also done on device characteristics for applications in potential applications in flexible substrates and systems, such as smart clothing, for example. The testing for this is executed using a polyimide substrate which is subsequently exposed to various bending conditions. The initial bending tests are done to find the maximum bending radius under which the devices would cease to operate properly, defined as follows; this is done by testing for device characteristics of a freshly fabricated array and then subjecting the array to various levels of bending and retesting to see how the characteristics of the device changed. Results for this line of testing yielded a maximum bending radius around 1.25 cm, with a slight increase in the response of the devices with a small bending radius 4.5–5.0 cm and a decrease towards the maximum bending radius (Figure 2c). The curve for this type of response with respect to bending radius fits closely with a negative second order polynomial function with the intercept set at the unbent level, with an R ² value of 0.997. One explanation for the initial increase in responsivity under these test conditions is that the uniaxial strain is shifting the peak of the band gap and thus the optical response to match the testing wavelength of 680 nm. The MoS₂ band gap is known to have a strong dependency on strain with a 45–70 meV redshift happening per degree of strain [12], [34], [35]. To shift the main excitonic peak it would be 27 meV or roughly 0.5 % strain, and to shift the secondary excitonic peak it would be a 216 meV or roughly 3–4 % strain. With a 3 mm-thick substrate and a 4.5 cm bending radius the strain induced on the MoS₂ is 3.3 % making it possible for the increase to be from the secondary excitonic peak shifted to be at the 680 nm testing wavelength. Furthermore, the reduced responsivity at mA/W levels is because testing was done with a laser operating around 10 μW approximately 2 orders of magnitude higher than the start of saturation that was seen previously. With the maximum bending radius established it is then used for stress testing over cycles between flat and the maximum bending radius, where the device characteristics are tracked over 100 cycles. The arrays experience relatively little change when tracked over the 100 cycles, having less than a 10 % change from the average responsivity after the initial bending. Throughout the bending cycles devices were also inspected for delamination of the MoS₂ flakes from the substrate and electrodes, across all 100 cycles no significant delamination was observed. This can be attributed to the relatively high Young’s modulus of MoS₂ in comparison with PI, which should have a lower breaking point from mechanical strain [36]. Allowing for the MoS₂ to bend with the substrate rather than delaminate, with a remaining possibility of delamination under more strenuous conditions. The results for responsivity under bending conditions are promising for usage in flexible systems, with special care being given to not exceed the maximum bending radius of the array.

Devices are also characterized under various preparation conditions, mainly by changing the annealing temperature and time for the devices. Both SiO₂ and PI substrates were tested, PI has a glass transition temperature around 250 °C leading to a maximum annealing temperature of 200 °C to avoid deformation. Previous research has shown that thermal annealing at low temperatures can help to create metallic phase transitions in the region between mechanically exfoliated TMDCs and electrode, thus leading to a lower contact resistance [37]. While other results have shown that Au-assisted exfoliation of monolayer material can result in little to benefit from thermal annealing, which could be caused from a minimal phase change seen with preheated gold electrodes, such is the case with a typical Au-assisted exfoliation method [37], [38]. Although previous demonstrations show minimization of resistance at 400 °C some improvements are shown at 200 °C and the surface roughness of the gold can help to increase transitions. Other researchers have shown that low temperature thermal annealing of MoS₂ based devices have 2 distinct regions where device improvements are <300 °C and a second region showing device degradation beyond 300 °C towards 400 °C in both photodetectors and FETs [38], [39], [40], [41]. Here, the devices see a 5× and 3× improvement on SiO₂ and PI substrates, respectively, the improvements are consistent with about half of the improvement seen at 100 °C anneal compared to the 200 °C anneal. These results suggest that annealing can help to improve the devices significantly and thus limiting the options for flexible substrates to those with high glass transition temperatures. Such substrates as PI and thin glass substrates with some potential on polyethylene naphthalate, although it has a glass transition temperature under 200 °C.

3.2 Photodetector array characterization and tuning

Photodetector array characterization is done by collecting all the device characteristics for a given 4 × 4 array and then comparing them with one another. I–V curves are collected per device from a range of −2 to 2 V of photoconductor bias in both dark and illuminated conditions (Figure 3a). This data and physical parameters are then used to find a normalized responsivity mapping of devices (Figure 3b). The mapping between devices shows significant variations with responses at 2 V and 680 nm ranging from 1 mA/W to 15.5 mA/W. Responsivities can roughly be clustered into three groups low performance (1–3.5 mA/W), middle performance (4–8 mA/W), and high performance (9–16 mA/W), while we again note, that experiments were carried out at laser powers where saturation occurred, else the responsivities are in the 0.× A/W range, see Figure 2b. No significant differences between these three groups can be established due to thickness, leading to a conclusion that innate resistive differences in the flakes and potentially contact resistance is leading to responsivity variations. Post-fabrication tuning for the array to generate a uniform photoabsorption condition and reduce the back-end burden on the digital signal processing (DSP) electronics, is done by changing the voltage bias on individual devices. The tuning level is set to be the lowest maximum responsivity of all devices, in this case devices 15 & 16 have 1.0 mA/W responses so that is set as the tuning level. The initial voltage for tuning was set at 1 V as the middle value for operational range of devices, from there the voltage is adjusted in 100 mV increments to reach the tuned responsivity level. Results for this methodology show bias voltages span the full range of 0.1 V–2 V, with the final voltage tunings resulting in −0.9 V to +1.0 V (Figure 3c). Reduction in voltage does not present a significant issue and can lead to reduced power consumption for the sensor, but voltage increases would require a method of amplification in a final system or potentially a local annealing method to further improve device performance. However, the maximally tuned up voltages still fall below the standard CMOS image sensor 3.3 V level, potentially leading to power savings overall using the MoS₂ detector array in an image sensor system. Final post-tuning response levels range from 0.9 mA/W to 1.5 mA/W showing much greater uniformity across the devices (Figure 3d). The bias voltage tuning condition also helps create a greater uniformity among device dark currents, moving from 50 % to 87.5 % agreement (Supplementary Material, Figure 4). We note here that it is possible to achieve 100 % agreement in the dark current but would come at the cost of responsivity agreement and an additional precision factor in bias voltage levels. Additionally non-uniformity in the dark current levels could require additional signal processing (e.g. DSP) to not trigger false positive readings on devices with higher dark currents. One potential avenue to further increase device uniformity is to use CVD grown MoS₂ rather than exfoliated material, which has been shown to have more uniform material characteristics and ultimately device performance [42]. The challenge with CVD growth is to find a suitable technique that would be amiable with growth on flexible substrates. Currently thermal CVD growth shows minimum temperatures around 500 °C, much higher than most flexible substrates can sustain [43], [44], [45]. However, recent advancements have shown potential in ALD and PECVD with lower growth temperatures (150–500 °C) for TMDCs creating potential for growth on some flexible substrates [46], [47], [48].

Figure 3:

Array variations and tuning. (a) Example of device I–V characteristics with operation with dark current and light current operated at 8.2 μW and 680 nm. Absolute value of the device current is taken to plot on a log scale graph. Inset shows device 7 on PI, the device under test, with a 10 μm scale bar. (b) As fabricated device characteristics for responsivities tested at 2 V, with colder colors representing lower device performance and warmer colors representing higher device performance. (c) Tuning to 1 V bias voltage applied to each device to increase uniformity for the full array. Warmer colors represent larger absolute value for the device tuning, while colder colors represent lower absolute value for tuning. (d) Array responsivities after applying voltage tuning from (c) minimal variation is seen between devices.

Figure 4:

2D material transfer process and system. (a) A flow chart for the 2D material transfer process showing automated and manual steps in the process where subroutines are marked with corresponding subsystems in (b). The process begins with a startup routine and then manually or automatically searching for suitable flakes that have been exfoliated on to a PDMS film, flakes must be manually chosen to be transferred to the given device region. Once the flake choice has been made the system will automatically do a rough alignment procedure and bring the flake to near contact with the device region, at which point the operator can choose to a manually fine alignment or to stamp the flake down for wider pitched devices. After the flake has been transferred the process can be repeated until all devices are created, at which point the system will run an automated end and rest routine. (b) 2D material transfer system and labelled subsystems. The first subsystem is an automated focus adjuster which is used in conjunction with the process monitor subsystem consisting of a camera and image processing software designed to find when images of the PDMS sheet are in focus. Utilizing these two subsystems and the third subsystem, the 2DM stage, flakes can automatically be searched for and located using edge detection and centroid localization algorithm. The final two subsystems control the transfer stamper and chip stages, respectively, both subsystems have 3 dimensional movements with the vertical axis of the transfer stamper using a sensitive piezo-electric system.