High synergy atomic layer etching of AlGaN/GaN with HBr and Ar

Atomic layer etching (ALE) is an iterative process by which monolayers of material can be removed sequentially by combining separate, self-limiting, reaction steps. This is achieved via a multistep process in which two key phases, surface modification followed by surface removal, limit the etching of material to individual atomic layers per cycle. ALE is of great interest for traditional semiconductors and has seen an influx in development over recent years.^1–6 As technology nodes increasingly demand sub-10 nm fidelity restraints, atomic-scale precision becomes a necessity.^6,7 The reactant of choice for directional ALE of Si has primarily been Cl₂ followed by Ar bombardment, with high directionality and minimal damage demonstrated.^6–11 Compound III-V materials present a more complex challenge to ALE development, compared to single element materials. Again, Cl₂ has been the reactant of choice in most directional ALE studies on materials such as GaAs^12,13 and GaN.^3,4 Based on the electrical characterization of surfaces exposed to ALE, several studies have shown minimal damage for both Si and III-V materials when compared to reactive ion etching.^8,14 More recently, HBr has been reported as a suitable reactant in directional ALE of both GaN and Ge,^5,15 demonstrating a less volatile response in comparison to Cl₂. Here, HBr was speculated to produce a thinner reaction layer than Cl₂ due to the larger diameter of HBr atoms while also producing less volatile etch products such as GaBr₃.⁵ A particular advantage of this lower volatility is reduced sidewall etching, observed for a HBr process when compared with Cl₂.⁵ 

A nondirectional ALE approach, sometimes referred to as “digital etching,” involves self-limiting surface modification via oxidation to form a reactive layer.^2,6,16 The oxidized layer is then removed, usually by a wet acid process such as HCl. This method is, thus, isotropic in nature and has seen significant development for use on III-V materials such as GaN and GaAs.^16,17 Compared to continuous wet etching, digital etching allows greater control of material removal by separating the process into two self-limiting steps. The process is usually labor intensive, however, and not in situ.

For field effect transistor devices based on AlGaN/GaN heterostructures, enhancement-mode operation is generally achieved by recessing the device gate closer to the 2DEG channel. A well calibrated directional ALE process with consistent and repeatable etch rate per cycle is well suited to achieving a gate recess with precise depth and minimal additional surface defects.¹⁸ Similarly, low resistance ohmic contacts can also be achieved by recessing them closer to the channel.¹⁹ In this work, we show experimental verification of an ALE process based on HBr + Ar gases for use on a commercially available AlGaN/GaN heterostructure.

Epitaxial AlGaN/GaN grown on a six-inch low resistivity Si wafer was purchased from NNT-AT Corp. Figure 1 shows the wafer heterostructure. The wafer was diced into 15 × 15 mm substrates and solvent cleaned in an ultrasonic bath using acetone and isopropyl. Immediately prior to patterning with photolithography, substrates were exposed to a HF solution for removal of the SiN top layer.

Substrates were then mounted to the center of individual Si carrier wafers with Santovac, to aid in thermal dissipation during processing. All etch work was carried out in an Oxford Instruments PlasmaPro System 100 ICP300 Cobra. At the start of each etch, prior to beginning the ALE cycle, a short 10 s, 45 W Ar plasma “breakthrough” step was used to prepare the surface and remove any residual SiN or native oxide. Postetch, the patterned resist was removed leaving a measurable etched step.

Figure 2 shows a complete ALE cycle consisting of an initial Reaction A, in which the surface is modified by HBr gas exposure (no plasma), followed by a purge step, and then Reaction B, in which the modified surface is removed by a low-power Ar plasma. This second reaction is intended to be self-limiting, stripping away only a monolayer of material exposed to HBr in Reaction A. HBr has been reported to form a thinner reaction layer on GaN, compared to other common ALE chemistries such as Cl₂⁵ while also producing the less volatile etch product, GaBr₃.^20,21 Thus, it was chosen for these experiments.

To achieve a self-limiting etch process, the “ALE window,” a low-power regime prior to sputtering, is established. This is determined by the sputtering energy threshold of the semiconductor surface^6,22 and experimentally is observed by gradually increasing the RF platen bias power of Reaction B (Ar plasma) during ALE cycles. We determined the ALE window for our process, as shown in Fig. 3, to be between 15 and 30 W. Here, a relatively consistent etch rate is observed prior to sputtering. As such, an RF platen bias power of 20 W, corresponding to a bias on the substrate of 120 V, was used in Reaction B step of our ALE experiments. It should be noted that for the initial trial data shown in Fig. 3, we did not mount the samples with Santovac. However, the data point at 20 W RF matches closely with our process results shown in Sec. III.

All measurements of depth postetch were measured by atomic force microscopy (AFM), using a Bruker Dimension Icon System. Step height was collected and averaged from four different locations on each sample, with an average standard deviation of 3.4% across measurements from each substrate. Surface roughness postetch was also measured by AFM, using a 5 μm scan size to ensure a good representation of each sample. Scans were taken from the center of etched regions, with average roughness (Ra) and the route mean squared (Rq) captured.

Etch depth as a function of ALE cycle is shown in Fig. 4(a), while Figs. 4(b)–4(e) show AFM scans of measured steps. Within the first ∼22 nm of material, good ALE behavior was observed with a etch rate of 0.81 Å/cycle for the AlGaN layer. However, the low etch rate suggests incomplete removal of monolayers per cycle and is potentially due to the Al content of the layer suppressing etch uniformity. Etching beyond the GaN layer transition depth, a sharp increase in etch rate can be seen, increasing to 2.64 Å/cycle. This removal rate for the GaN layer is close to the 3.1 Å lattice parameter of GaN (Ref. 23) and suggests good uniformity. Similar to ALE of SiN using H₂/Ar (where H serves to remove N),²⁴ in GaN/AlGaN, we suspect N reacts to form NH₃. For GaN exposed to HBr, the etch product GaBr₃ has been observed.^5,6 The etch product GaBr₃ has been reported to possess a higher sublimation point compared to GaCl₃.^20,21

At the transition, the interface between AlGaN and GaN is also a thin, 1 nm AlN layer included for enhanced charge mobility in the GaN channel.²⁵ Etch rates extracted for both AlGaN and GaN layers appear linear and largely unaffected by this thin AlN layer, though its presence should still be considered when interpreting the presented results.

Also plotted in Fig. 4(a) is the etch depth for 240 and 400 cycles of Ar only exposure. This was done by way of a modified version of the ALE recipe shown in Fig. 2, with the HBr component of Reaction A removed. Isolated, the Ar plasma step of Reaction B showed a very low etch depth of ∼3 nm after 400 cycles. In addition, 240 cycles of the ALE process with only HBr exposure (Reaction A only, Ar plasma removed) showed no measurable etching of the surface. This result is comparable to that observed elsewhere.¹ 

For the steps plotted in Figs. 4(b)–4(e), it should be noted that the AFM technique, while highly accurate at measuring height, is not capable of accurately representing sidewall slope. This is due to the pyramidal shape of the AFM tip, resulting in the apex of the tip being unable to physically contact the sidewall of vertical features. This effect becomes more pronounced as the step height increases.

The ALE synergy test is a measure of efficiency for the process and quantifies the contribution of individual steps. It is often used in determining the degree to which an ALE process approaches ideal behavior, calculated by the following:

where Adsorption_Rate is the etch rate per cycle for the HBr exposure step (Reaction A), Activation_Rate is the etch rate per cycle for the Ar plasma exposure step (Reaction B), and ALE_rate is the etch rate per cycle for both reactions combined. Achieving 100% ALE synergy would mean both Reactions A and B, in isolation, result in no surface material removed. In practice, the active plasma exposure of Reaction B typically results in some small degree of etching. Synergy results for the etch per cycle data plotted in Fig. 4 are shown in Fig. 5. Both AlGaN and GaN ALE processes showed very high synergy of 91%–97%, largely due to the low parasitic impact of Reaction B alone. This synergy result is similar to that reported by Kauppinen et al.,³ and significantly higher than some reported elsewhere,^1,4 though many published ALE papers do not discuss synergy results.

Roughness analysis of the surface after successive etch cycles is shown in Fig. 6 up to 400 cycles. To ensure a representative measure of roughness, a 5 μm AFM scan was taken from the center of each etch region. Results showed an increase in surface Rq while etching into the AlGaN layer (from 0.7 to 1.1 nm), caused by what appears to be micromasking effects. These surface peaks measured around 5 nm in height and dissipate gradually as the etch moves into the GaN layer, suggesting the effect to be AlGaN specific. The same micromasking of AlGaN was also observed by Ohba et al. for a Cl₂ based ALE process on GaN/AlGaN.¹ Based on XPS analysis of surfaces after etching, they proposed that the micromasking was caused by the formation of nonvolatile AlO. A solution was found through the introduction of BCl₃ to the chamber, acting as a scavenger to form BClO.²⁶ After this modification, the surface Rq of their AlGaN layers returned to pre-etch levels. We, therefore, expect that similar optimization of our process, with the inclusion of BCl₃ to Reaction A (HBr surface modification step), would remedy the increased AlGaN roughness. Without this modification, the observed surface roughness for our process shown here resulted in a Rq increase of 0.4 nm after 100 cycles (AlGaN layer) and 0.2 nm after 400 cycles (GaN layer) when compared to initial starting values. While fair comparisons of roughness data between reported literature are difficult, due to variations in substrates, etch processes, and measurement procedures such as scan size, an increase in surface Rq of 0.6 nm was reported for ALE of GaN after 200 cycles using Cl₂.³ 

In summary, a process of atomic layer etching of AlGaN/GaN heterostructures is detailed. Precise control of the etch process was achieved using HBr and Ar, with process synergy over 91%. Analysis of surfaces postetch showed increased roughness of AlGaN layers, potentially caused by nonvolatile AlO formation. However, roughness decreased to values similar to starting levels after etching into the GaN layer. We expect the adoption of atomic layer etching using less volatile gases, such as HBr, to be highly successful in the production of high-performance GaN field effect devices.

This work was supported by Grant No. EP/V026127/1. The authors wish to acknowledge discussions with Igic and Faramehr as well as the staff and facilities of the James Watt Nanofabrication Centre.

The authors have no conflicts to disclose.

The data that support the findings of this study are available from the corresponding author upon reasonable request.