Improving the Manufacturability of Low-Temperature GaN Ohmic Contact by Blocking the Fluorine Ion Injection

Stabilizing the CMOS-compatible low-temperature Au-free GaN Ohmic contact is a critical work that determines the performance and yield of GaN power HEMTs in mass production. The instability of this contact has been puzzling the industry and academia for years. In this work, an overlooked factor, fluorine injection, is unambiguously verified to widely exist during dielectric etching and can easily destroy the low-temperature GaN Ohmic contact formation. The injection depth is confirmed to be over 30 nm with a fluorine peak concentration of 1023 at/cm3 in vicinity of the surface. Traditional method of partial AlGaN recessing with a pretty tiny processing window is proven unfriendly for production and vulnerable to the fluorine injection. Two methods to get rid of the fluorine are proposed. The first one is to over-etch the AlGaN barrier to the GaN channel to fully remove the fluorine ions. The second is to deposit an etch-stop blocking layer of Al2O3, which is also compatible with CMOS process.


I. INTRODUCTION
GaN HEMTs offers great perspectives for high-performance power devices for many applications such as renewable energy, industrial electronics, and electric vehicles, thanks to the high electron mobility, high critical electric field, and high frequency [1], [2], [3], [4], [5], [6], although most GaN power devices are today still employed in consumer electronics.To date, several aspects that hinder the widespread applications of GaN HEMTs still exists, for instance, the relatively high cost, low yield, and nonformalized reliability and processing.
To lower the cost and promote mass production, highproductivity CMOS-compatible and Au-free process is indispensable.However, most eminent source/drain metallization schemes are based on Au-containing metal stacks assisted by rapid thermal annealing (RTA) under a very high temperature of above 800 • C [7], [8], [9].This is anyway unsuitable because of the high cost.Thus, some other techniques, for instance, Ti/Al/Ni/TiN [10], Ti/Al [11], Ti/Al/W [12], and TiAl alloy [13] are proposed.Low Ohmic contact resistance R C of < 1 •mm can be obtained at the risk of high annealing temperature, which is harmful to the self-aligned gate-first process and the passivation dielectrics [14].Plus, high-temperature treatment may also trigger some reliability problems [15], [16].
Low-temperature metallization scheme is then achieved by Ta/Al metal stack that is unfortunately not compatible with CMOS process [17].Further optimized Au-free lowtemperature metallization scheme is thus proposed, which focuses on the widely used Ti/Al stack in microelectronics industry.It is believed that plenty of donor-like N vacancies in AlGaN or GaN can be generated so that to form the Ohmic contact [18], [19].Firrincieli et al. found that the Ti/Al ratio of 0.05 is crucial to reduce R C of the Au-free lowtemperature Ti/Al/TiN contact to 0.62 ± 0.06 •mm [20].Pre-ohmic recess etching of AlGaN barrier [21], [22] was also proven effective to reduce the R C .However, the strict thickness of the residual AlGaN barrier leads to a pretty

II. EXPERIMENTS
The AlGaN/GaN heterostructure wafer used in this paper was grown on a 6-inch Si substrate by metal organic chemical vapor deposition system (MOCVD).The epitaxy stack consists of 5 nm GaN cap layer, 20 nm Al 0.25 Ga 0.75 N barrier layer, 250 nm GaN channel layer, ∼5 μm GaN buffer layer, and 200 nm AlN nucleation layer, from top to bottom, as shown in Fig. 1.Hall mobility and 2DEG sheet resistance of the AlGaN/GaN heterostructure reach 1800 cm 2 / V•s and 380 / , respectively.200 nm SiO 2 was first deposited on the GaN wafer using plasma enhanced chemical vapor deposition (PECVD) to passivate the top surface.Then the Ohmic contact windows were patterned by lithography, after which the SiO 2 was etched off by CHF 3 -based RIE with an etching rate of about 22 nm/min.Afterwards, part of the AlGaN barrier layer was recessed by BCl 3 /Cl 2 -based inductively coupled plasma (ICP) dry etching, and the power of the ICP source and RF generator were set to 150 and 25 W, respectively, with a BCl 3 /Cl 2 flow rate of 8/20 sccm and a pressure of 20 mTorr.The etching rate is tuned to as low as 0.9 nm/s.Notably, to investigate the impact of RIE and ICP, various etching time and depth were investigated.Next, the photoresist was stripped and the samples were subjected to a wet treatment in 18% HCl to remove the surface oxides.Finally, Ti/Al (10 nm/200 nm) was deposited using e-beam evaporation as the Ohmic contact metal, followed by a final RTAfor 60 s at 550 • C in the ambient of N 2 .All samples have the same epitaxial structure and fabrication process, except for variations in passivation stack, RIE and ICP etching time.Detailed sample structure and processing flow are summarized in TABLE 1.The microstructures of the samples were characterized using scanning electron microscopy (SEM) and transmission electron microscopy (TEM).After fabrication, the Ohmic contacts were subjected to electrical characterization with a circular transmission line model (CTLM) by Keysight B1500A, where the test structure is plotted in Fig. 2. In order to eliminate the parasitic effect of probe resistance, a four-point probe method was adopted for current testing.

A. IMPACT OF RIE OVER-ETCHING ON OHMIC CONTACT
In reality, certain level of over-etching of SiO 2 or other passivation dielectric layer is always obligatory due to the inhomogeneity of the dielectric thickness and RIE etching.Therefore, the first group of comparative experiments come with etching the same 200 nm SiO 2 with three different RIE  etching time of 10, 11, and 12 minutes, corresponding to the over-etching time of about 1, 2, and 3 minutes, respectively, as shown in Fig. 3(a).The AlGaN barrier layer was then partially recessed for 25 s by ICP, leading to a ∼2.5 nm AlGaN barrier layer retained, as shown by the SEM image in Fig. 3(b).These three samples are denoted as sample A10, A11, and A12.Fig. 4 demonstrates the I-V characteristics of the fabricated CTLM on the three samples.Schottky but not Ohmic contacts are formed, and the drain current decreases with the RIE over-etching time, indicating the RIE etching can probably damage the Ohmic contact.

B. VERIFICATION OF THE FLUORINE INJECTION
In order to mitigate the possible interference by the SiO 2 dielectric itself, we further conducted RIE etching on the blanket GaN wafer without any dielectric on top for 0, 2, and 4 minutes, followed by the subsequent 25-s AlGaN recessing, metal evaporation, and RTA.These three samples are denoted as sample B0, B2, and B4.Fig. 5 illustrates •cm −2 , as shown in Fig. 6.Obviously, the current keeps on degrading from 624 to 283, and finally to 132 mA/mm as the RIE etching time increases from 0 to 4 minutes.This clear trend unambiguously proves that the RIE over-etching does jeopardize the Ohmic contact formation.
To further verify the existence of fluorine ions in the epitaxial layer, secondary ion mass spectroscopy (SIMS) profiles of the fluorine were then probed as shown in Fig. 7.The fluorine peak concentration reaches 3×10 22 at/cm 3 after RIE etching of 4 minutes.A potential barrier is therefore lifted up between the metal and the AlGaN barrier, thus destroying the Ohmic contact formation [23].•cm −2 .According to the SIMS profile in Fig. 7, it can be easily concluded that the impact of fluorine injection can be sufficiently removed by over-recessing the AlGaN barrier.

D. SUPPRESSING THE IMPACT OF FLUORINE INJECTION BY AN ETCH-STOP BLOCKING LAYER
It is also noticed that some other teams favor the nonrecessed [24] or partially-recessed [22] low-temperature Ohmic contact.Therefore, another 5-nm Al 2 O 3 etch-stop layer is proposed here to block the fluorine injection.In this experiment, the 5-nm Al 2 O 3 was first deposited by plasma-enhanced atomic layer deposition (PEALD) on the GaN wafer, denoted as sample D5, followed by 200 nm SiO 2 deposition and then RIE etching for 12 minutes with an over-etching time of ∼ 3 minutes.Before the Ohmic metal Ti/Al deposition, hydrochloric acid wet etching was employed to remove the Al 2 O 3 and 25-s AlGaN recessing by ICP, as shown in Fig. 10(a).The microstructure of the

IV. DISCUSSION
Temperature-dependent I-V characteristics of the sample A11 with 11-min RIE etching and subsequent 25-s AlGaN recessing are plotted in Fig. 12(a).The I-V curves are clearly divided into two parts by a point of intersection.That is, from 0 to 2 V, the current increases with the temperature due to the thermally enhanced field emission [23]; from 2 V, the current drops due to the mobility degradation at higher temperatures.This phenomenon is basic for the Schottky barrier diode [25], indicating that a potential barrier has been  The schematic contact mechanisms and the influence of fluorine injection are depicted in Fig. 13.The residual AlGaN thickness after recessing is crucial.As seen from Fig. 13(a) to (c), the key to guarantee the formation of Ohmic contact is to ensure the distance between the contact metal and the 2DEG.Previous works have claimed to retain part of the AlGaN barrier to keep the 2DEG under the contact metal, so as to give a better contact.This method is difficult to be employed in mass production because the strict thickness results in a pretty tiny processing window.Moreover, this Ohmic contact can be destroyed by the fluorine injection as shown in Fig. 13(d), where the detailed mechanism is plotted by the inset of Fig. 12.Therefore, over-recessing is more practical, where the 2DEG can be connected to the metal through the side wall as in Fig. 13(c).Even if the contact suffers the fluorine injection, over-etching can easily etch off theses negative ions, which is exactly the case of sample C40.

V. CONCLUSION
Impact of fluorine injection on the CMOS-compatible lowtemperature Au-free GaN Ohmic contact has been explored in this work.The existence and drawback of fluorine injection during the Ohmic contact window opening by RIE etching are unambiguously verified.Two effective methods to suppress this impact were proposed.The first is to fully recess the AlGaN barrier even to the GaN channel layer to physically remove the injected fluorine.The second method is to deposit a Al 2 O 3 etch-stop blocking layer to fully block the fluorine injection from the root.These two methods are both compatible with the mass production of GaN HEMTs.Moreover, the confirmation of the fluorine injection successfully explains the long-standing puzzle of the terrible repeatability of GaN low-temperature Ohmic contact, which is of great significance for boosting the manufacturability of GaN power HEMTs.

FIGURE 2 .TABLE 1 .
FIGURE 2. Photograph of the fabricated CTLM test structure.TABLE 1. Summary of the processed samples in this work.Sample structure process Ohmic contact?A10

FIGURE 3 .
FIGURE 3. (a) Schematic structure of sample A10, A11, and A12, and (b) SEM image of sample A10 with 10-min RIE etching on the SiO 2 and subsequent 25-s AlGaN recessing by ICP.

FIGURE 4 .
FIGURE 4. I-V characteristics of the sample A10, A11, and A12, corresponding to RIE etching of the 200 nm SiO 2 for 10, 11, and 12 minutes and subsequent 25-s AlGaN recessing by ICP.

3 )FIGURE 7 .
FIGURE 7. SIMS distribution profiles of the fluorine ions in the epitaxial layer induced by RIE etching of sample B0, B2, and B4, and distribution profiles of Al and Ga.C. SUPPRESSING THE IMPACT OF FLUORINE INJECTION BY ALGAN RECESSINGNext, solutions to mitigate the drawback of fluorine injection during RIE etching is proposed.The first method is to physically remove these ions by recessing the AlGaN barrier and even the GaN channel with ICP etching.After opening the Ohmic contact window by RIE etching of the 200 nm SiO 2 for 12 minutes, AlGaN/GaN recessing by ICP etching was then performed for 20, 25, and 40 seconds, corresponding to etching depth of 18, 22.5, and 36 nm, as shown in Fig.8(a).These three samples are denoted as sample C20, C25, and C40, respectively.TEM images in Fig. 8(b)-(d) depict the detailed recessed AlGaN/GaN heterostructures after various ICP etching.Over-etching of the GaN channel of Sample C40 can be clearly observed, which may more thoroughly remove the fluorine.From Fig. 9(a) we can see the current does improve with the ICP etching time, indicating the Ohmic contact can be indeed restored by over-recessing the AlGaN barrier.The Ohmic contact parameters extracted from Fig. 9(b) shows sample C40 features a R C of 1.05 •cm and ρ c of 3.48×10 −5•cm −2 .According to the SIMS profile in Fig.7, it can be easily concluded that the impact of fluorine injection can be sufficiently removed by over-recessing the AlGaN barrier.

FIGURE 8 .
FIGURE 8. (a) Schematic structure of sample C20, C25, and C40, and TEM images of the contact structures of sample (b) C20, (c) C25, and (d) C40 with different ICP etching time of 20 s, 25 s, and 40 s, respectively.sample D5 was checked by the SEM image in Fig. 10(b), where the Al 2 O 3 /SiO 2 passivation stack can be observed.R C of 1.45 •cm and ρ c of 6.27×10 −5 •cm −2 are achieved for sample D5 with the 5-nm Al 2 O 3 etch-stop blocking layer, as shown in Fig. 11.

FIGURE 9 .FIGURE 10 .FIGURE 11 .
FIGURE 9. (a) I-V characteristics of the sample C20, C25, and C40 after ICP etching for 20 s, 25 s, and 40 s, respectively, and (b) linear fitting of total resistance versus spacing of sample C40.