Sub-bandgap photon-assisted electron trapping and detrapping in AlGaN/GaN heterostructure field-effect transistors

We have investigated photon-assisted trapping and detrapping of electrons injected from the gate under negative bias in a heterostructure field-effect transistor (HFET). The electron injection rate from the gate was found to be dramatically affected by sub-bandgap laser illumination. The trapped electrons reduced the two-dimensional electron gas (2DEG) density at the AlGaN/GaN heterointerface but could also be emitted from their trap states by sub-bandgap photons, leading to a recovery of 2DEG density. The trapping and detrapping dynamics were found to be strongly dependent on the wavelength and focal position of the laser, as well as the gate bias stress time prior to illumination of the HFET. Applying this phenomenon of trapping and detrapping assisted by sub-bandgap photons, red, green, and purple lasers were used to demonstrate photo-assisted dynamic switching operations by manipulation of trapped carriers at the surface of an AlGaN/GaN HFET. A physical model based on band diagrams, explaining the trapping and detrapping behavior of electrons, has been presented.


Introduction
III-Nitride heterostructure field effect transistors (HFETs) have been widely researched in the past few decades, in part due to their applications in high-power microwave devices [1][2][3][4][5][6]. A significant problem faced by these HFETs, especially in the early phases of their development, was the tunneling of electrons under a high electric field from the gate into surface states (from a combination of negative voltage bias at the gate and large positive voltage at the drain), which resulted in the problem of 'current slump' or 'current collapse' that significantly degraded the performance of these devices at microwave frequencies [7][8][9][10][11][12]. The trapped electrons on the surface have the effect of partially depleting the two-dimensional electron gas (2DEG), which is formed at the AlGaN/GaN interface without any intentional doping due to the strong polarization properties of III-nitrides [2,3]. Obviously, the drain current of the HFET gets strongly impacted by such trapped charges and remains reduced until they are neutralized completely. This problem has been largely mitigated in advanced devices by the usage of surface passivation (with AlN, SiN x , or SiO 2 ) [8][9][10][11][12][13][14][15] and gate field plates [16][17][18], which has resulted in excellent microwave performance [10,[19][20][21]. Furthermore, surface Kelvin probe microscopy (SKPM) has been used previously to map out the changes in surface potential of an AlGaN/GaN HFET after applied bias stress and has shown that at the edge of the gate on the drain side is where the greatest change in potential occurs [8]. This heavily supports that the most active states, which have the greatest effect on the channel electrons, are those at the bare surface near the gate edge.
Although generally considered a nuisance for microwave devices, one of the interesting aspects of these surface traps is their trap activation energy and consequent long emission time constant, which has consequences in localizing trapped charges for a long time at the surface [8,13]. Spatial localization (of trapped charges) down to 1 μm has been reported previously (studied through surface barrier measurements using SKPM) where the trapped charges were found to be localized for days and even weeks [13]. The trapped charges would spread out with time (accelerated at higher temperatures), but exposure to super-bandgap (> GaN bandgap) ultraviolet (UV) light neutralized them immediately. The long storage times of the trapped charges provide an interesting platform for studying 'memory effects' similar to the floating gate memory formed with Si metal oxide semiconductor field-effect transistor (MOSFET) devices [22,23].
Although UV illumination nullifies the trapped charges (photogenerated holes migrate toward the surface, assisted by the built-in electric field, and neutralize trapped electrons), the influence of sub-bandgap illumination on the dynamics of surface trapping and detrapping can be significant and more interesting. Typically, the effect of sub-bandgap illumination using green/red light has been studied in the interest of mitigating the effect of trapped charges through photo-assisted emission of the electrons from surface traps without generating holes that can have unintentional side effects on device performance (i.e., enhanced gate leakage current). Although many studies have reported on the emission of trapped electrons using sub-bandgap light [14,24,25], only a few studies have reported on the impact of sub-bandgap illumination on the gate tunneling current in an HFET [15,26]. In fact, we did not find any previous report carefully considering the competing phenomena of electron injection from the gate (and into surface states) and trapped electron emission from surface states in AlGaN/GaN HFET structures, both assisted by sub-bandgap photons. In this work, we have performed a systematic study on the photon-assisted emission of electrons from the gate Schottky contact into surface states, as well as the simultaneous emission of electrons from surface states. The influence of gate bias, wavelength, and relative intensity/location of the laser beam (used as illumination source) on trapping and detrapping processes in III-Nitride HFETs has been demonstrated. We would like to mention that traps located directly beneath the gate region should have been relatively unaffected by laser illumination, because the gate metal should reflect incident photons, preventing them from reaching the traps. Thus, the emphasis of this paper will be on trap states at the bare surface, which we believe to be the main contributors to our observed phenomena. A physical model, based on band diagrams, explaining our results has also been included.

Experimental methods
The GaN-based HFET devices used in our experiments were fabricated using III-nitride epilayers, grown on a Si (111) wafer purchased from DOWA Semiconductor Akita Co., Ltd., Japan. The HFETs were fabricated at the base of microcantilever beams, which have nominal dimensions of 250 × 50 × 1.3 μm [27]. Aside from their placement on microcantilever beams, the HFETs are similar in all aspects to those fabricated conventionally on solid substrates. The HFET layer structure is shown in figure 1(a) and consists of a 675 μm Si (111) substrate layer, a 300 nm buffer layer, a 1 μm i-GaN active layer, a 20 nm barrier layer of intrinsic Al 0.25 Ga 0.75 N, and finally a cap layer of 3 nm i-GaN. The mesa structure (14 × 34 μm) for the HFET at the microcantilever base was defined through photolithographic patterning followed by etching of the AlGaN layer using an inductively coupled plasma (ICP) based dry etch process. Then, the cantilever outline was patterned, and the GaN layer was etched. The gate was positioned at the center, between the source and drain, with the gate-drain (and gatesource) spacing ∼1.5 μm. The metal stack for the source/drain ohmic contacts consisted of Ti (20 nm)/Al (100 nm)/Ti (45 nm)/Au (55 nm) layers deposited via electron beam (e-beam) evaporation and then finished by rapid thermal annealing at 800°C for 1 min Ni (25 nm)/Au (375 nm) Schottky gate contacts were then deposited, followed by Ti (25 nm)/Au (200 nm) probe contacts using e-beam evaporation. Finally, throughwafer Si etching was performed using a Bosch process to release the microcantilevers. A more detailed description of the fabrication process has been presented elsewhere [27]. We want to mention here that the release process was not found to affect the HFET performance to any significant extent.
To study the trapping/detrapping effects of sub-bandgap light on the HFET surface, we used 635 nm, 520 nm, and 405 nm wavelength lasers with rated powers of 125 mW, 25 mW, and 30 mW, respectively. The beams emitted from these lasers were Gaussian, which enabled us to adjust the laser intensity (incident on the gate and surface) by slightly adjusting the laser focal position. We also incorporated the fluorescent room lights (which typically emit some UV components [15,28]) into some of our HFET response measurements for additional comparisons. A ×4 neutral density filter (NDF) was used to reduce laser intensity by ∼75%, when necessary. Figure 1(b) shows close-up images of some measurement-ready devices. A processed chip containing several cantilevers was glued to a 28-pin dual inline package (DIP) chip carrier and the gate, drain, and source contact pads were wire bonded to designated pins. Figure 1(c) gives the schematic of our experimental setup used to measure changes in HFET channel resistance from laser exposure at different levels of gate bias. A laser was focused and pulsed at two different locations: Position 1, slightly in front of the HFET (with the less intense areas of the Gaussian distributed beam still overlapping the HFET), and Position 2, directly on the HFET. Before turning on the laser, a negative voltage was applied to the gate, and a constant current was supplied from drain to source (I DS ). A precision source/measure unit (SMU, model B2912A) from Keysight Technologies was used to supply the constant current (10 μA) while simultaneously measuring the drain-source voltage (V DS ). Changes in HFET channel resistance were tracked by monitoring changes in V DS [directly proportional to changes in drain-source resistance (R DS ) because I DS was held constant]. A lock-in amplifier (SR850, Stanford Research Systems) was used to apply the HFET gate bias and supply the control signal for pulsing the laser (pulsed at 100 Hz to reduce damage to the HFET from sustained exposure).

Results and discussion
3.1. Device characterization Two similar HFETs fabricated on the same wafer were used throughout our experiments, referred to as 'Device 1' and 'Device 2'. Basic electrical characteristics, i.e., I DS -V DS and I DS -V GS curves, for both HFETs are shown in figure 2. The device characteristics were used to calculate the 2DEG mobility in the linear regime (μ 2DEG ) using the equation: where L ch is channel length (11.5 μm), W ch is channel width (14 μm), C' eff is the effective capacitance per unit area between the gate and the channel, R sheet is channel sheet resistance, q is elementary electron charge, and V DS is the drain-source voltage, which was taken as 0.5 V to correspond to the value of transconductance (g m ) obtained from the I DS -V GS measurements.  [2], and the C' eff was calculated from the above formula as 361.52 nF cm −2 . Figure 2(a) gives the I DS -V DS characteristics for Device 1, where V GS was decreased from 0 V down to −4.2 V in steps of −0.1 V. The maximum saturation drain-source current (I DSS ) was found (from the V GS = 0 V curve) to be 2.97 mA, which reduced monotonically and very uniformly with application of more negative gate bias. Figure 2(b) shows an I DS -V GS curve for Device 1 with V DS set at 0.5 V, from which the threshold voltage (V T , arbitrarily defined as the value of V GS where the I DS is 1000 times smaller than I DSS ) was determined to be ∼−3.82 V. The transconductance g m was obtained to be 287.6 μS, from the slope of the linear region of the curve [marked by the black dotted line in figure 2(b)]. Using this value of g m , along with the values of C' eff , V DS , L ch , and W ch , the μ 2DEG was found from equation (1) as 1307 cm 2 V−s, which agrees with previously reported values (>1000 cm 2 V−s [3]). The n s was calculated from equation (2) as 4.56 × 10 12 cm −2 , which is lower than typically expected from theoretical considerations (∼10 13 cm −2 [2]), likely due to high density of trapped electrons from extensive usage/biasing of the device, but comparable to previously reported values for these devices [29]. Using the same methods as above, μ 2DEG and n s values for Device 2 were extracted from I-V measurements, which are shown in figures 2(c) and (d).  (1) and (2), we obtained the mobility and carrier density as μ 2DEG = 990.7 cm 2 V−s and n s = 4.49 × 10 12 cm −2 , respectively. We find that the 2DEG density for Device 2 is very comparable to that of Device 1, while the carrier mobility is significantly lower, which can be attributed to material and device fabrication related non-uniformities.

Surface trapping of gate-ejected electrons
As mentioned above, under the influence of a negative gate bias and a positive drain bias, electrons can tunnel from the gate Schottky metal into the surface trap states between the gate and drain. At higher negative gate bias values, the probability of field-emission becomes greater, leading to a higher rate of electron trapping on the surface. These trapped electrons then reduce the 2DEG density at the AlGaN/GaN interface, which increases the R DS . Since we maintained the drain current at a constant value of 10 μA, V DS increased proportionally with R DS . V DS measurements from Device 1 under dark conditions are given in figure 3, which show the V DS to increase in steps corresponding to the different levels of V GS applied (starting from 0 V and systematically decreasing to −3.3 V). We note that immediately following the sharp jumps due to V GS change are sloped rises, which can be attributed to the gradual trapping of electrons ejected from the gate. As the V GS magnitude was increased, the trapping rate also increased, causing the V DS to rise faster, which was quite dramatic around V GS = −3.3 V. The sharp decline in V DS at the end of the final step (V GS = −3.3 V) was caused intentionally by setting the V GS back to 0 V. Application of further negative gate bias would cause a sharper increase of drain voltage, ultimately leading to a runaway condition and saturation of V DS at the SMU compliance voltage (set at 10 V to protect the HFET). This runaway condition will be discussed in detail later.
The inset of figure 3 shows the total changes in R DS and n s from their initial (V GS = 0 V ) levels. From the plot, the total increase in V DS is ∼0.24 V, with R DS increasing proportionally by 24 kΩ and n s decreasing by 3.12 × 10 12 cm −2 . The magnitude of Δn s must equal the change in trapped surface charge by the condition of charge neutrality. So, with n s decreasing by 3.12 × 10 12 cm −2 , we can conclude that the density of trapped surface charge has increased by the same amount.

Photon enhanced electron ejection, surface trapping and detrapping
While the surface electron trapping behavior under a high gate-drain electric field has been widely reported in the literature [8][9][10][11][12], there are only a few studies investigating the enhancement of this trapping behavior under sub-bandgap light (laser) exposure. Our investigation on this began with an accidental discovery while sweeping the focus of a 635 nm laser beam across the surface of the gold-coated chip carrier [passing very briefly across Position 1, labeled in figure 1(c)], which resulted in a significant spike in V DS . Further exploring the cause of the spike, we held the laser location constant in Position 1 and recorded the V DS responses at six different levels of V GS (varying from −2.6 V to −3.35 V). After each gate bias adjustment, the laser was turned on, resulting in the transients shown in figure 4, which are compared with a V DS measurement obtained without laser exposure under the same biasing conditions. Figure 4(a) shows the first three bias levels with arrows labeling the points where V GS was adjusted, and where the 635 nm laser was turned on/off. As indicated in figure 4, noticeable jumps in V DS were observed from Device 1 under laser exposure, with the magnitude of the jumps monotonically increasing with the magnitude of gate bias. Turning the laser off caused V DS to reduce gradually back to its initial step level. Figure 4(b) shows the V DS responses recorded at three additional gate voltages where we see more pronounced laser-induced transient behavior corresponding to higher negative bias levels. For V GS = −3.3 V and −3.35 V, we found that the V DS greatly increased under initial laser exposure, immediately followed by a strong transient reduction (with the laser still on). This transient reduction behavior is also present in the responses at lower step levels but is much less pronounced. The inset of figure 4(b) shows V DS plotted in log scale, which shows that for V GS = −3.35 V, V DS briefly reached the SMU compliance level of 10 V before decaying.
It should be noted that the fluorescent room lights were intentionally left on while collecting both data sets shown in figure 4 to facilitate a clear comparison between the laser on/off cases. The room lights provided the benefit of stabilizing the V DS responses beyond each gate bias induced jump, giving us noticeably flatter-looking steps (compared to those in figure 3) and clear contrast between the two measurement sets. Additionally, at each point where the laser was turned off, the room lights quickly reset the V DS back to the initial step level, allowing a smooth transition into the next gate bias induced jump. These stabilizing ambient lighting effects can be attributed to the UVA (315-400 nm) and UVB (280-315 nm) radiation emitted from the fluorescent bulbs [15,28]. The UV photons, having energy comparable to or larger than the bandgaps of GaN and Al 0.25 Ga 0.75 N (365 nm and 302 nm, respectively [30][31][32]), can generate electron-hole pairs within the AlGaN and GaN layers, which are then swept by the built-in electric field across the AlGaN barrier layer. This results in the electrons moving to the channel to enhance the 2DEG density and the holes moving to the surface to neutralize the trapped charges. For the 'no laser' measurements in figure 4 (shown in blue), the stability of V DS at each bias level was likely maintained due to an established equilibrium between the rate of surface trapping and the rate of electron-hole pair generation caused by fluorescent room light exposure. In contrast, figure 3 measurements were taken in complete darkness, and as a result these V DS transients have additional sloped rises caused by the gradual accumulation of trapped surface charge near the gate, as mentioned previously. With the HFET kept entirely in the dark, there were no substantial means by which the rate of surface trapping could be mitigated, so when V GS was adjusted to higher magnitudes, the trapping rate kept increasing monotonically with time.
As mentioned above, the sharp initial increases in V DS upon exposure to the laser (more prominent at the higher bias levels) can be attributed to significant enhancement in electron emission from the gate and, consequently, higher electron trapping at the surface states. However, with very similar photon flux on both the gate edge and the bare surface (almost equally covered by the large beam diameter of the laser), electron emission from surface states should also be enhanced by the laser illumination. Thus, the transient reduction in V DS following the initial steep rise was likely caused by photo-assisted detrapping of electrons from the filled surface states, which has been widely studied and reported [24][25][26]. After the initial high rate of electron injection into the surface states (indicated by the initial jump in V DS ), the detrapping rate increased considerably, even exceeding the initial trapping rate. However, the detrapping rate gradually decreased with time until an equilibrium was established between the two rates. This is indicated by the monotonically reducing V DS transient following the spike, which eventually levels out.
In contrast to the above, we also studied the laser-assisted trapping dynamics with the room lights off, and the results are shown in figure 5. This measurement was taken from Device 2 with the laser held in Position 1. We once again observed the initial quick rise in V DS caused by the application of gate voltage (−3.3 V), which was then followed by a sloped increasing transient indicative of surface trapping (this time not suppressed by room light). When the red laser was turned on, just as in figure 4, there was a sharp rise in V DS , which eventually stabilized, indicating equilibrium between the trapping and detrapping rates. However, unlike in figure 4 when the red laser was switched off, we observed V DS to monotonically increase over the span of an hour, indicating a nearly constant rate of trapping at the surface, which was unmitigated due to the absence of ambient room light. Finally, when the room lights were turned on, the V DS reduced rather sharply to the initial level, clearly underlining the role of ambient light in stabilizing the carrier dynamics in AlGaN/GaN HFETs. We want to emphasize here that in the absence of ambient lighting, the laser itself can stabilize the V DS after the initial jump, as indicated by the stable V DS maintained under prolonged laser exposure. This indication of equilibrium between the rates of trapping and detrapping further establishes the role of the laser in facilitating both mechanisms simultaneously. Further discussion on the trapping/detrapping dynamics is presented in conjunction with our physical model.

Trapping and detrapping dependence on laser position and wavelength
We mentioned earlier that the laser position plays a critical role in the trapping and detrapping dynamics. The beams emitted from these lasers are Gaussian, so between laser Positions 1 and 2 [see figure 1(c)], there is a difference in optical intensity incident on the HFET, which influences electron injection from the gate and detrapping from the surface states. In addition, we have observed a strong dependence of the electron trapping and detrapping dynamics on the laser wavelength and the initial gate bias stress time (which leads to different densities of surface-trapped charges at the instant of laser exposure). To further investigate these effects, under dark conditions we recorded sets of V DS transients from Device 2, exposing it to the 635 nm and 520 nm lasers (separately). The position of the laser was changed between Positions 1 and 2 in-between measurements. Two sets of these position/wavelength comparisons were recorded: one for a pre-illumination gate bias stress period of 200 s and the other for 800 s. Figure 6(a) compares the first set of V DS transients from the 635 nm and 520 nm lasers, which were turned on after 200 s of gate bias stress. Before illumination from the 635 nm laser in Position 1, the applied gate bias caused an increasing V DS transient. When the laser was switched on, there was an immediate jump in V DS , as observed in previous data sets (figures 4 and 5), but unlike in figures 4 and 5, no equilibrium V DS level was reached, and instead we saw V DS gradually increase after the initial jump. On the other hand, the 635 nm Position 2 data shows a slow reduction in V DS following the initial rise, similar to that in figure 4(a) (red curves). From the 520 nm laser exposure, the Position 1 response was similar to that of the 635 nm laser, although the magnitude and rate of increase were significantly lower. In Position 2, the V DS decreased in contrast to the 635 nm exposure and monotonically kept reducing, signifying a continuous detrapping of electrons.
When the initial gate bias application time was increased from 200 s to 800 s (V GS kept at −3.3 V), the responses changed significantly. The response from 635 nm laser exposure in Position 1 caused a similar rise in V DS initially, however, V DS reached equilibrium much quicker compared to the 200 s case. The response at Position 2 was different, showing a spike in V DS , followed by a much more pronounced decay transient compared to that in figure 6(a), reducing the V DS below the level when the laser was switched on. The responses for the 520 nm laser were also significantly different from those in figure 6(a), with both Position 1 and 2 exposures resulting in a clear downward trend in V DS as a function of time, indicating a higher rate of detrapping compared to trapping.
We want to mention here that using a filter to reduce the optical intensity affected the V DS value as expected. The inset of figure 6(b) shows the V DS response as 520 nm illumination was reduced in intensity by ∼75% using a neutral density filter. As the filter was put in place, V DS increased as expected, since the reduced illumination intensity should cause a greater reduction of the detrapping rate compared to the trapping rate, leading to an increase in V DS . The trapping and detrapping processes as facilitated by the laser illumination are explained with the help of band diagrams and surface trap models below.

Physical model
To explain the observed V DS transients, we propose a physical model to explain the influence of sub-bandgap optical illumination on the charge trapping and detrapping effects in the HFET. Figure 7(a) shows the trapping of electrons under the application of a negative gate bias and sub-bandgap illumination. Under negative bias, the primary mechanism under which electrons eject from the gate is thermionic-field emission through the thin forbidden gap, as shown in figure 7(a) [33]. These charges get trapped at the interband states at the surface or subsurface region and deplete the 2DEG density at the AlGaN/GaN interface [see figure 7(b)], which increases the V DS due to the constant current condition of our setup. Under optical illumination, the electrons can also emit over the barrier to reach the surface states via internal photoemission; 635 nm and 520 nm correspond to photon energies of 1.95 eV and 2.38 eV, respectively, which are much larger than the typical Ni/GaN Schottkybarrier height of ∼0.9-1 eV [34][35][36]. Both of these mechanisms are shown schematically in figure 7(a). At a larger negative gate bias, the reverse bias current increases, and more trapping happens near the edge of the gate, which results in more pronounced transients (both with and without laser illumination) as seen in figures 3 and 4. The trapped charge density can reach sufficient levels to almost completely deplete the 2DEG, increasing the R DS and V DS to very high levels [ figure 4(b) inset]. We believe the mechanism of thermionic-field emission is made even more pronounced than typically observed, due to our device biasing setup, which keeps I DS constant while trapped electrons deplete the 2DEG. Because of this constant current condition, as R DS increases with trapped charge density, so does Joule heating (I 2 R loss) in the depleted channel region. The associated temperature rise can cause increased thermionic-field emission in a reverse biased Schottky barrier, which would cause more electron trapping, depleting the channel further, with resultant increase in I 2 R heating, and so on. This is the thermal runaway condition mentioned earlier, which is possible if the HFET is kept in the dark under sufficient bias stress. We see the beginning of this condition in figures 3 and 5 with the upward sloped transients after the application of the gate bias. With enough time allowed under sufficiently high gate bias, we've observed these sloped transients gradually become exponential, eventually reaching the 10 V SMU compliance. Figure 4 shows this behavior suppressed by the fluorescent ambient lighting, as indicated by the decay in V DS after the laser is turned off. Even more notable however, is that figures 5 and 6 demonstrate that sub-bandgap laser exposure can also suppress this behavior, as discussed below.
Optical illumination can also enhance the opposing mechanism, assisting the emission of trapped charges from the surface/subsurface states, as shown schematically in figure 7. Due to the large beam diameters of our lasers, it is very likely that the processes depicted in figures 7(a) and (c) coincided at variable rates. This is also supported by our data, most notably, by the flatness in V DS established in figure 5 for the duration of laser exposure (after the initial rise). This flatness from figure 5 is especially significant because it occurred under dark conditions, which suggests that laser-assisted detrapping can be sufficient to enforce an equilibrium condition. Also notable is that V DS resumed its upward trend in figure 5 after the laser was turned off, indicating that the trapping behavior was being suppressed for the duration of 635 nm exposure.
As mentioned, the flatness indicates the establishment of an equilibrium condition. In figure 5 this occurs shortly after the initial high rate of electron injection from the gate upon turning on the laser (indicated by the spike in V DS ). As more electrons populate the surface very close to the gate, a negative space charge region is formed, which opposes further injection of gate electrons [37,38], consequently reducing the trapping rate. At the same time, with more electrons populating the surface, the rate of electron emission from the surface increases. Equilibrium is established when the negative space charge develops sufficiently, such that the gate electron emission rate (and subsequent trapping rate) is reduced to match the detrapping rate. The critical density of trapped surface charge constituting the negative space charge region determines the net increase observed in V DS , which remains constant (with an even flow of electrons entering and leaving the surface) so long as the laser intensity, focal location, and level of gate bias are maintained.
Also important in controlling the electron dynamics (injection from the gate or emission from trap states) are the laser intensity and the density of electrons in the region of laser focus. This is illustrated by the responses shown with the filter being on and off [figure 6(b) inset]. Additionally, and more subtly, this is exhibited by the laser location dependence of the trapping and detrapping processes observed in figure 6. While the laser is focused on the HFET (Position 2), the gate edge and HFET surface both receive the maximum intensity of the laser light, which maximizes both the electron injection from the gate, and emission from surface states. When the laser is focused somewhat away from the HFET (Position 1), the intensity is reduced and the ejection rate from both the gate and surface states reduces. Ultimately what determines the accumulated equilibrium density of surface charge resulting from laser illumination is how the initial ejection rate from the gate compares with that from traps. The figure 6 data suggests that the latter is affected more considerably than the former when laser intensity is changed. Taking the figure 6(a) 635 nm transients for example, one can notice that the increase in V DS from illumination in Position 1 is higher than from Position 2. This means that in spite of the laser intensity being reduced (in Position 1), the initial net injection of carriers onto the surface is greater. A reasonable explanation for this would be that the electron ejection rate from surface states has reduced more than from the gate, between Positions 2 and 1, which is likely because of the much higher density of electrons in the gate metal compared to the surface.
On the other hand, if there is a sufficient density of trapped carriers already present on the surface due to prolonged bias stress, it is possible that their emission can be facilitated at a higher rate than that from the gate, causing the equilibrium density to be reached much quicker. This is because the emission from the traps is strongly dependent on the magnitude of the trapped carrier density. The higher the trapped carrier density, the higher the probability of electron emission upon exposure to light. In addition, the trapped electrons oppose further ejection of electrons from the gate, reducing the trapping rate. This is clearly demonstrated in the responses to 520 nm and 635 nm exposure after 200 s and 800 s of charge trapping (enabled by negative gate bias) shown in figure 6. The higher levels of trapped charges, after 800 s of gate biasing, for both wavelengths, resulted in transient reductions of V DS after the initial sharp rise, which then approached equilibrium much quicker than the figure 6(a) transients. This behavior was especially noticeable under 520 nm illumination. The difference between the responses from the 635 and 520 nm lasers seen in figure 6 is likely due to the energy levels of the surface traps. While both lasers can emit electrons from the gate similarly (as their photon energies are much higher than the gate Schottky barrier), they will impact the trapped charge emission differently due to the different energy depth distribution of the surface/subsurface traps. This is shown schematically in figure 7(c), where only shallower traps emit electrons upon 635 nm laser exposure, while both shallower and deeper traps can emit upon 520 nm laser exposure, as observed experimentally in figure 6. Thus, 520 nm exposure results in a higher rate of photo-assisted detrapping, resulting in a lower equilibrium V DS compared to 635 nm exposure. The deepest traps can respond to 405 nm laser exposure which can also create electron-hole pairs (EHPs) due to its photon energy (3.06 eV) being quite close to the bandgap of GaN (3.42 eV). This photon absorption due to band tailing [32] creates photogenerated holes that can reach the surface and neutralize the trapped electrons. Thus, this laser was always observed to lower V DS very quickly, showing relatively little position dependence.

Photo-assisted switching operations
Taking advantage of the resetting capability (through removal of trapped charges) of the 520 and 405 nm lasers, we exposed the HFET to sequences of 635/405 nm illumination and 635/520 nm illumination to demonstrate the possibility of photo-assisted switching ('memory') operations. In these switching operations, the 'high' state in V DS was achieved by maintaining only the 635 nm laser illumination, while the 'low' state (reduction in V DS ) was caused by maintaining only the 520/405 nm laser illumination.
For the first switching sequence shown in figure 8(a), the 635 nm laser was focused on the HFET (Position 2), while the 405 nm laser was focused on Position 1 with the gate voltage maintained at −3.45 V. Upon initial exposure to only the 635 nm laser, the V DS exhibited a sharp jump followed by a rising V DS transient caused by enhancement in electron trapping, while subsequent exposure to the 405 nm laser (with the 635 nm turned off) reset V DS almost completely. Figure 8(b) shows similar switching responses, where the 'high' state (due to electron trapping) was achieved again with 635 nm exposure in Position 2, and the 'low' state (from electron detrapping) was achieved with 520 nm illumination. To optimize detrapping performance, the 520 nm laser was focused directly on the HFET. Both the 635/405 nm and 635/520 nm combinations resulted in quite uniform rise and fall magnitudes over multiple cycles indicating stable device operation. We note that the V DS level at the 'low' state caused by the 405 nm laser is slightly lower than that caused by the 520 nm laser. This is expected since The 635 nm laser, held in Position 2, is used to accumulate negative surface charge, setting V DS high. The 520 nm laser, also held in Position 2, is used to deplete negative surface charge, resetting V DS low. the creation of electron-hole pairs by the former can lead to more efficient reduction of surface traps through recombination with holes. From one individual pulse, the rise time constant for V DS was determined to be ∼1-2 s, while the fall time constant was found to be ∼200 ms. Although the switching power (and time constants) required for 'memory' operations in these devices are much higher than those necessary for practical applications, it still provides opportunities for further exploitation of the phenomenon to develop devices for dynamic memory applications. To this end, miniaturization of the devices and periodic reading (rather than continuous reading by passing constant drain current) of the high/low states could significantly reduce the operational power requirement.

Conclusions
In conclusion, we have demonstrated the strong influence of sub-bandgap laser illumination on the trapping and detrapping dynamics in AlGaN/GaN HFETs. Trapping/detrapping dependence on laser position was observed with detrapping being more prominent in Position 2 (with the HFET under higher optical intensity) compared to Position 1. The trapping/detrapping behavior was also found to be a strong function of laser wavelength, with the red laser facilitating relatively higher rates of electron injection from the gate compared to emission of trapped electrons. The green laser seemed to favor the latter process more than the red, which can be attributed to its interaction with the deeper traps within the bandgap. In contrast to the red and green lasers, the purple laser was found to produce an overwhelming detrapping response with little positional dependence, as it caused photo-generation of electron-hole pairs, where the holes recombined with trapped electrons at the surface. The carrier dynamics were also found to be dependent on the density of trapped surface charge prior to illumination, which is a function of the duration of negative gate bias. Photon-assisted dynamic switching operations between high and low drain voltage states were also demonstrated by alternately switching the red with the green or purple lasers. A physical model based on band diagrams has been proposed to explain the experimentally observed photon-assisted carrier dynamics.