Lithographic Processes

20

the next steps. Here you see the alignment of the second mask. The initial markers for the next steps will be defined here. The letters in every initial marker represent the name of the step; for example, "I" stands for isolation mask and "M" stands for metallization, which is P-metal deposition.

21

seconds. Therefore etch rate is 300 nm/min. We will use the same solution also for the mesa etch.

Figure 2.9. a) Schematic of facet after cap layer etch. b) Top view of the device after cap layer etch taken by optical microscopy with 10x lens.

Chemically etching everywhere results in a rough surface which increases the adhesion of the silicon nitride layer. We use silicon nitride as the electrical isolation layer not to inject current outside the waveguide.We will discuss the details about the insulator layer in the p-metal window opening part. Cap layer etching is also decreasing the catastrophic optical mirror damage by protecting the facet from the current leakage. We will also explain it in the next chapter.

Mesa Etch

Mesa step is where we define the ridges. It creates the refractive index difference between the ridge and outside. The waveguide should have a higher refractive index than the surrounding. As we discussed earlier, we have two waveguides close to each other to make mode coupling possible between them.

22

As we mentioned in the cap layer section, an essential point after mesa etching decreases the catastrophic optical mirror damage effect. We have a gap between the end of the cap layer etch, and the mesa etch. When we cleave the laser bar to define facets, we still have an etched part where the current does not flow. Thus, it limits possible leakage towards the laser facet. This leaking current would result in damaging the facet and decrease the lifetime of the laser device. Since the current does not flow to the facet, it will not damage the facet, as shown in Figure 2.10 . I label this gap as un-pumped region.

Therefore, the current will directly flow through the waveguide as shown in Figure 2.10 .

Figure 2.10. a) Sechematic of facet after cap layer etch. b) Sechematic of facet after mesa etches c) Top view of the device after cap layer etch taken by optical microscopy d) Top view of the device after mesa etch taken by optical microscopy.

The challenging part starts with the lithography. We set the spacing between the waveguides to 2 µm in the initial design in this step because we planned to do H+

Implantation. It adds extra steps to the device fabrication. We use a hard mask to protect waveguides from implantation. While implanting, they accelerate the atoms and hit the

23

substrate; then atom sits in a suitable place in the crystal structure. The hard mask protects the waveguide so that particles cannot pass through the mask because it is thick enough to hold them. As you can imagine, this extra step will result in an undercut that increases the spacing between the waveguides. To keep the final spacing around 2 µm, we design the mask to 1.3 µm. It becomes 1.7 µm after hard mask etching, and reaches 2 µm at the end of substrate etching.

In a standard semiconductor laser device fabrication, the etching method is chemical etching. The main difference between the chemical and physical etching is the anisotropy. For anisotropic etching, we use physical etching, and for isotropic etching, we use chemical etching. Etching type affects the wall shapes of the ridge. In the P-metal deposition step, we need to have a continuous metal layer on the walls. Chemical etching provides this positive angled wall shape for a continuous metal layer. Negative angled mesa walls will result in discontinued top metal contact, which makes the plating step impossible.

Chemical etching is essential for laser fabrication. Based on a systematic study published on GaAs etching [32], [33], we aimed to obtain the same wall shape for my epitaxial design. We successfully repeated the results for GaAs, but it didn't give good results for the laser structure we employed for the device fabrication. The results are summarized in Figure 2.11 . As seen in Figure 2.11., chemical etching strongly depends on the crystal orientation and substrate. We have two different crystal orientations in this experiment, perpendicular and parallel orientation, defined according to major flat. The results for the 0-deg-off GaAs wafer with HCl: H2O2:DI (1:4:40) satisfy our expectations, but the same solution does not give a similar wall shape with the 0-deg-off laser sample.

24

According to these results, the undercut is more than expectations as it expands the spacing width to 3 µm. Then, we focused on anisotropic etch optimization with ICP to have less undercut. To overcome discontinues in the top metal contact layer, we will make the top contact thicker by electrochemical gold deposition. If the metal layer is thicker than my ridge height, the contact metal will be continuous as desired.

Figure 2.11. Three different etch solutions, parallel and perpendicular, represent the crystal orientation.

We start optimizing the anisotropic etching for silicon nitride with RIE and laser sample etching with ICP. We will discuss the results of anisotropic silicon nitride etching in the next chapter. Here we will show the final results of anisotropic substrate (mostly GaAs and AlGaAs) etching with ICP. We used the recipes from the literature to optimize for our structure [34]–[38]. After obtaining a working recipe, we modify the gas ratios to make them suitable for our case. Figure 2.12 presents a schematic and the etching results of four different recipes and the red circle in the schematics shows SEM picture area. The first three are with the same gases, and the difference between them is the gas ratios.

25

Different ratios change the undercut but not in a linear way. The spacing width for the recipes is 1.463 µm, 1.571 µm, 1.473 µm, and 1.968 µm, respectively. Changing the gasses is directly affecting the etch rate. Etch rates for the recipes are 1.4 µm/min, 1.6 µm/min, 1.4 µm/min and 1.9 µm/min respectively. According to the results, recipe b and d look suitable but recipe b has some roughness on the walls. Then, we decided to continue with recipe d.

Figure 2.12. a) Schematic of structure b) BCl3:Cl2 (5:15) c) BCl3:Cl2 (10:15) d) BCl3:Cl2 (15:15) e) BCl3:Cl2:Ar:N2

(10:10:10:4.5).

After physical etching with ICP, we continue with chemical etching for 20 seconds to obtain smooth wall morphology. We represent the SEM pictures of defined mesa steps after etching in Figure 2.13. The SEM pictures from the facet and optical images of top surface are presented in Figure 2.13.

Figure 2.13. a) SEM picture of the facet b) Optical microscope image of the surface after etching the nitride film.

26

Etch Process Development for H+ Implantation

In the project, the first idea was to do hydrogen implantation for electrical isolation of the waveguides. This section discusses the details of the hard mask required for the implantation process. We need a hard mask to protect some regions from implanted hydrogen ions during the implantation process.

We employed the mesa etch step mask for patterning the hard mask, so it does not add a step to the lithography process. We deposit silicon nitride to the surface with PECVD, then pattern it with PR. We continue with etching silicon nitride with RIE, etching the substrate with ICP, and then chemical etching. To deposit 800 nm silicon nitride, we used a PECVD recipe with Silane (SiH4,%2): NH3 (180:45 sccm) at 250 C° and 1 Torr. Then, it is patterned by ~500 nm thick PR. We then use an RIE recipe with CHF3:O2 (100:5 sccm), 20 µbar, 71-W power to etch the silicon nitride. This recipe gives excellent results as presented in Figure 2.14. The SEM picture shows the facet profile after mesa etch with the hard mask on top, where the final spacing is around 2 µm.

Figure 2.14. SEM picture after etching the nitride and substrate.

Then the silane gas we are using for depositing silicon nitride was depleted.

Replacing gas took around six months and we focused on optimizing new recipes using

27

another PECVD machine to deposit silicon nitride. We first deposit with the same recipe I used before in the new machine. Then we tried to etch it with the same RIE recipe, but it didn't give the same results. In Figure 2.15. you see four pictures from two different recipes; a-b) are the same recipe that I used before, c-d) are the SF6:O2 (20:5 sccm) 20µbar 71 W. Figure 2.15 a-c) shows the silicon nitride residues in the opening part after the RIE process. I added 10-sec BOE etching to remove the leftover silicon nitride residues. In Figure 2.15 b-d) you can see the clean version of the opening after BOE etching.

Figure 2.15. Spacing between the waveguides after silicon nitride etching a) CHF3:O2 (100:5 sccm) 20 µbar 71 watt b) CHF3:O2 (100:5 sccm) 20 µbar 71 watt + 10-sec BOE etching c) SF6:O2 (20:5 sccm) 20 µbar 71 watt d) SF6:O2 (20:5 sccm) 20 µbar 71 watt + 10-sec BOE.

The spacing width is ~2 µm in Figure 2.15.a), and ~3 µm in b). Since there are still residues, we need to do BOE etch for ten or twenty more seconds, increasing the spacing.

In

Table 2.3, all the results from our RIE experiments are summarized. In the meantime, RIE broke down, and we had to move to ICP for etching silicon nitride. It has various gas options, and etch rate is higher than RIE.

We did optimization with ICP to etch silicon nitride anisotropically. In this part of the thesis, we will summarize the results of these experiments. There are few things to optimize with ICP etching: ICP Power, Bias Power, frequency of ICP power, and gas ratios. First, we need to find a recipe that etches the silicon nitride we deposit. We tried some recipes from literature and colleagues. The recipes from the literature generally do

28

not lead to the results presented due to the difference of machine architectures. However, the recipes from the colleagues provide better results. ICP machine has two power supplies;

ICP Power and Bias Power. ICP Power provides higher density plasma, and it increases the etch rate. If you apply zero ICP Power, ICP becomes RIE because RIE has only Bias Power. Previously we tried the same etching recipe we used in RIE, with zero ICP power.

We didn't get the same results that we used to get on the same silicon nitride layer in Figure 2.16. Comparing the results from this etching experiment, a) represents the results from RIE etching b) ICP etching. Although we set the same values, the results are not similar.

In Figure 2.16, RIE etched all the silicon nitride but ICP caused undercut with a ratio of almost one.

Table 2.3. We named the recipes for the people who will use the same machine within our facility; UNAM and ARL are the names of two different clean rooms within our facility. For example, the difference between UNAM He and UNAM N2 recipes is the carrier in the Silane gas, UNAM He has %2 Silane and %98 He, UNAM N2 %2 Silane %98 N2, these carriers shouldn't change the quality of silicon nitride because the only difference is the carrier gas. I already give the details of the etching recipes in the previous discussion.

Machine Name of the recipe

Deposition Rate (nm per minute)

ETCH RATE (nm per minute)

BOE CHF3:O2 100:5

CH4F3:O2 60:40

SF6:O2 20:5

PECVD 2

UNAM

He - - 8.625 - -

UNAM

N2 21.34

~1400 14 (40 min etch with PR)

7.75

20 (10 min etching)

~300 8.75 (40 min etch WO PR) 16.67 (30 min etching)

~180 9.34 (30 min etch WO PR) 13.4 (50 min etching)

OLD PECVD

ARL He 7.27

- 11.25 - -

6.82

ARL N2 12.7 ~180 21 - -

ARL

SiO2 34

Hot: 450

- - -

R.T.:

150

29

The goal of these experiments is to have a repeatable working recipe with maximum selectivity. The PR thickness is 460 nm, and the silicon nitride thickness is 700 nm as required for the implantation process. Therefore, selectivity should be higher than 1.52. In Figure 2.17, you can see the effect of ICP power frequency. There are two frequency options for the ICP Power; 380 kHz and 13.56 mHz. Higher frequency provides higher plasma density. As you can see in Figure 2.17, surprisingly, higher density plasma etches less PR than lower density plasma, and undercut is also less compared to the lower density plasma.

Figure 2.16: a) Etching recipe I used with RIE CHF3:O2 (100:5 sccm) 20 µbar 71 watt b) Etching recipe I used with ICP CHF3:O2 (100:5 sccm), 15 mTorr, ICP Power: 0 W, Bias Power: 71 W. The pressure might seem different but when we convert µbar to torr they are the same.

Since plasma is etching the substrate physically by hitting the surface atoms with ions and ejecting them, ICP is categorized among the physical etching methods. However, due to the interaction between the gasses and the substrate, chemical etching is occurring.

These minor interactions may increase due to the ratio of etching gasses. The results of this chemical etching are the undercut. Therefore, we can decrease the undercut by changing the gas ratio. In Figure 2.19. you see the three different gas ratios and subsequent undercut

30

values. We have two other gases, CHF4 and O2 ratio, in the recipes, representing the flow rate of CHF4/O2. In Figure 2.19 ratio is 5, 1.34, and 1, respectively, where undercut values are on the bottom of the pictures. They are not similar to our expectation. They do not change linearly due to the huge difference between the ratios. For example, by setting the ratio to 2 instead of 5, we would see the linear change in the undercut values. Because when we increase the volume of gas inside the chamber, other unknown interactions might happen. It is better the put values close to each other. It seems that from Figure 2.19 b-c), a higher gas ratio lowers the undercut due to the lower interaction between the gasses and the substrate.

Figure 2.17. a) Etching results of the recipe with low-frequency ICP power b) Etching results of the recipe with high-frequency ICP power.

Figure 2.18. SEM pictures from the facet view after following recipes a) Bias power: 300 watt, ICP power:

30 watt, CF4:O2 (100:20 sccm), Pressure: 40 mTorr b) Bias power: 100 watt, ICP power: 30 watt, CF4:O2

31

(100:20 sccm), Pressure: 40 mTorr c) Bias power: 70 watt, ICP power: 30 watt, CF4:O2 (20:15 sccm), Pressure: 5 mTorr d) Bias power: 70 watt, ICP power: 30 watt, CF4:O2 (20:15 sccm), Pressure: 40 mTorr.

Figure 2.19. SEM pictures from the facet view after following recipes a) Bias power: 70 watt, ICP power:

30 watt, CF4:O2 (100:20 sccm), Pressure: 40 mTorr b) Bias power: 70 watt, ICP power: 30 watt, CF4:O2 (20:15 sccm), Pressure: 40 mTorr c) Bias power: 100 watt, ICP power: 30 watt, CF4:O2 (20:20 sccm), Pressure: 40 mTorr

We showed some of the results from different experiments; we end up with this recipe which might work for our case. Bias power: 400 watt, ICP power: 100 watt, CF4:CHF3 (100:20 sccm), Pressure: 40 mTorr. After deciding on the recipe, we should check if it is repeatable. For this, we tried the same recipe for a different amount of time.

In Figure 2.20. you can see the results from different etch times from 1 to 3.5 min.

According to the results, we don't have a repeatable recipe; even the etch rates change from one process to another. We realized that inside the chamber etching process is not homogenous. If we don't place the sample in the same location every time, we have different results.

32

Figure 2.20: SEM pictures from the facet view after best recipe (Bias power: 400 watt, ICP power: 100 watt, CF4:CHF3 (20:3 sccm), Pressure: 5 mTorr) with different etch duration a) 1-minute b) 2-minute c) 2.5-minute d) 3-2.5-minute e) 3.5-2.5-minute.

We implemented a two-step implantation process. First step is 3E15/cm² at 50keV then second step is 2E15/cm² at 20keV. The implant energy determines the depth for waveguide isolation as illustrated in Figure 2.21. The H+ implanted areas protect the lasing part so that we can isolate waveguides from each other.

Figure 2.21. The facet view of the implantation process.

P-Metal Window Opening

Now we deposit the isolation layer make a path for the current to follow through the waveguide. On top of the ridges, we etch some part as a window to ensure that the current will flow through there. We represent the structure after metal window opening, both schematic, and optical microscope image in Figure 2.22. From those openings, current will flow through the waveguide.

33

Figure 2.22.: a) Facet view of the structure after window opening. b) optical microscopy image from the top after window opening, purple areas are silicon nitride, and white areas are the window opening.

We have the most critical alignment in this step. If it is shifted more than 1.25 µm due to the misalignment, window opening will be on the spacing to pump the area we want to isolate. To protect from this, in the new mask design, we increased the separation between the windows as much as possible. We shift them in the opposite direction through the edges of the waveguides. Another thing here is, we were using BOE for etching the silicon nitride for opening windows. Silicon nitride thickness is 150 nm, and BOE etch rate is too high, see

Table 2.3. due to this high etch rate, controlling the undercut was difficult. To protect from opening more expansive windows than ridges, we did the window opening with RIE to have anisotropic walls. In Figure 2.23, you can see the results from the anisotropic window opening.

34

Figure 2.23. a) Schematic after metal window opening etch b) SEM picture of facet view after metal window opening etch with RIE, nitride layer is comparable to the schematic.

Metal Contacts Deposition

The semiconductor laser is a diode that allows current to flow in only one direction. For electrical pumping, metal layers and subsequent wire bonding are needed. We deposit two metal layers as top contact (p-metal) and bottom contact (n-metal).

We deposit gold (Au) due to its high conductivity. The problem with the gold is that it doesn't stick to the GaAs surface directly. Since our substrate is GaAs, we start deposition with 20 nm Ti (Titanium) by an e-beam evaporator. The titanium layer sticks to the GaAs and the other metal layers stick to the Titanium (Ti). Deposition continues with the Pt (Platinum) layer by e-beam evaporation to block gold diffusion during the annealing process Finally, we deposit a 100 nm gold layer by e-beam evaporation. The directional coating of e-beam evaporation makes the subsequent lift-off process possible.

In a laser chip, as you can see in Figure 2.6. There are eleven lasers in one bar and 6 bars in total in one chip. To separate lasers in bars and waveguides in PTS lasers from each other, we remove the metal layer in-between them by doing a lift-off. We do image reversal lithography to have a PR structure suitable for lift-off. In Figure 2.24, you can see the SEM picture showing the PR profile after image reversal lithography. Positive

35

angled walls and directional coating make lift-off possible. The photoresist thickness is around 1.4 µm, and the deposited metals are 145 nm in total,. SEM picture shows the facet view of the PR after development in Figure 2.25.a). The length of our laser is 4 mm with spacing around 4 µm. The critical lift-off will be here, and the PR profile of this process is in Figure 2.25.b). The photoresist in Figure 2.26. b) has a 4 mm length and ~4 µm width in this step. Considering the aspect ratio, the liftoff process is difficult. To achieve this, we decrease the vacuum level to 10^-6 mBar and also prevented photoresist from baking during metal deposition.

Figure 2.24. a) The facet view of the PR profile after image reversal lithography b) Side view of the PR profile that will sit on the spacing between the waveguides, after image reversal lithography.

Decreasing resistivity of the top contact layer is increasing the laser performance.

One option to reduce the top layer resistivity is to increase the gold layer thickness. Since we deposit gold with e-beam evaporation, increasing the thickness of the gold layer makes the lift-off process difficult. After some thickness, making lift-off is impossible due to the chamber's increasing temperature. Temperature increase results in the hard-baking photoresist and cause issues in the lift-off process. Therefore, we increased the gold thickness to 1 µm level by electrochemical deposition. It is crucial to isolate the waveguides

36

from each other in these two top metal contact layer deposition steps. Figure 2.25 illustrates a schematic, and top view of after p-metal deposition and gold plating. As seen from the results, they are separate.

Figure 2.25. a) Facet view of of final structure schematic b) Top view optical microscopy picture after p-metal deposition (145 nm gold) c) Top view optical microscopy image after electroplating (~1µm gold).

Isolation and Cleave

Isolation of the laser chips from each other is essential due to the current spreading. As you can see in Figure 2.6 we have eleven lasers next to each other in every bar. Since we are cleaving them from the bars, we should isolate them from each other; otherwise, they all lase when we pump one of them.

We use chemical etching to isolate the lasers. In our structure, the substrate starts after ~5 µm from the top, so 10 µm etch depth is enough for isolation.

We also create 10 µm deep cleaving marks during this step to guide the diamond tip while cleaving.

37