Etching of SiO2 in BOE Spring 1999 EE527 Amy Skrobis Xiaorong Xiong General Safety Requirements: HF solutions are considered both toxic and corrosive. Use of Trionic chemical gloves, safety glasses and a fume hood are required as a minimum. Check the condition of the Trionic gloves by inserting a N2 gun into the cuff and sealing around the gun; watch for leaks/pinholes. If any are observed, discard the gloves. Etch only in the blue polypropylene tanks in the Non-CMOS fume hood. Do not etch in chemical glassware. Do not etch outside of the fume hood. The wet bench should have dry surfaces on completion of etching. Any drips or spills are to be treated as a chemical spill. Use the HF Spill Kit to clean up after any BOE or HF spills; do not use Chem Wipes. For further information on cleanup of chemical spills or correct disposal procedures, reference course pak “Techniques Lab. Handbook”, Vol. 1, Section [7]. If BOE does contact the skin or eyes, if there’s even a suspicion it has, rinse with copious amounts of water immediately. Seek medical attention. For further information regarding safety practices, please refer back to course pak “Techniques Lab Handbook”, Section [6].

General Remarks regarding the etching of SiO2/Quartz/Glass The etching of SiO2 films undergoes the following reaction [1-4]: SiO2 (s) + 6 HF (l) → H2SiF6 + 2 H2O If etched with a buffered HF (containing NH4F), two species are present: HF → H+ + Fk1= 10-3 HF + F → HF2 k2=10-1 BOE (or BHF) etches display a sensitivity to agitation ([4], p.649) and are driven by the latter reaction. Unbuffered HF etchants by comparison dissociate less readily and produce slower etch rates. Additionally, unbuffered HF solutions are powerful

resist penetrants and failure due to pinholing and delamination are more likely than with buffered oxide etches [4]. Undercutting of the resist layer is a common problem to oxide etching. It is for this reason that measures must be taken to insure the best possible resist adhesion. The oxide surface should be clean and free of water (“singe” step). Typical recommended conditions for such a dehydration step are 30 minutes convection bake at 200-250 C; however, total dehydration doesn’t occur until 750 C --- and is re-absorbed upon cooling. [3]. Surface treatments like HMDS (hexamethyldisilazane) dehydrate a surface by reacting with silanol groups (OHgroups bonded to the Si-substrate) to form non-polar trimethysilyl groups. Adhesion of the resist is improved by lowering the surface tension of the substrate allowing resist to flow or “wet” the substrate rather than “beading up” [4]. One test of the efficacy of the priming step is to place droplets of water onto the primed wafer and measure the contact angle. After HMDS priming the water should bead, as the surface has become hydrophobic, just as DQ-impregnated resist is at this point in processing. This layer also should prevent penetration of polar liquids, such as developer and BOE etchants. HMDS, however, is typically applied in a separate chamber due to the ammonia and the hexamethyldisiloxane vapors produced. This side reaction can cause diazocoupling and silyation of the phenolic ester, producing an insoluble resist [[3] And Lecture Handout, Positive Photoresist – DQ Side Reactions (1) and (2)]. Therefore, care should be exercised within the SCS spin bowl of the Microfabrication Lab to achieve optimum results.

Characterization of Exposure Energy Seven oxide wafers (thickness ~5000 A) produced via thermal oxidation were obtained from Tai Chen. Six received a one minute 90 C “singe”. Less than 3 minutes of time elapsed between “singe” and coat for all wafers. A full puddle (wafer fully covered) of AZ1512 was then spun on using the following recipe: Step 1: 100 RPM 5 seconds Step 2: 500 RPM 10 seconds Step 3: 4000 RPM 20 seconds Wafers were then prebaked for 45 seconds at 90 C. Three were then selected and measured on a Nanospec Nanometrics AFT located at AlliedSignal using program #10, positive resist on silicon. Program was referenced to the seventh bare oxide wafer. The following thicknesses (in A) were measured at five sites on the wafers, with site three being in the location of the flat [Fig. 1]:

1 4 Thickness 5 2 Figure [1]: Measurement Sites 3

Top 1 18905 18520 18716

Right 2 18877 18513 18760

Bottom 3 18918 18392 18978

Left 4 18727 18364 18929

Center 5 19464 19103 19596

Average Thickness: 18850 A Standard Deviation: 340 A

19800 19600 19400 19200 19000 Wafer1

18800

Wafer2 Wafer3

18600 18400 18200 18000 17800 17600 Top 1

Bottom 3

Center 5

Figure 2. Wafer Thickness By Site

This is thicker than anticipated as AZ1512 should produce around a 1.2 micron film. Added thickness should be beneficial during the wet etch, making a throughresist attack less likely. The wafers were exposed using EE527 Metal-2 mask on the Oriel aligner. In all instances of aligner use, the lamp was given in excess of a one-hour warm-up time. The following exposure times were then run: 2, 3, 5, 7.5, 10, 15 seconds.

The wafers then received 45 second develop (kept as a constant) in AZ300 MIF developer. The wafers were returned to AlliedSignal to again be measured under the Nanospec to determine residual resist and create a contrast curve (exposure vs ln (remaining thickness)-1. This was not possible as the lowest exposure chosen had clean patterns. Therefore, a visual inspection was done. The 2 second exposure was best, with alignment cross spaces (“opens” in the resist) retaining 90o corners. The narrowest of these cross features measured 10.84 microns using a BioRad linewidth measurement system. By contrast, higher exposures produced rounding of the features and a general recession or pulling back of the resist from the pattern; the same cross feature with a 15 second exposure measured 18.09 microns on the BioRad and was relatively less well defined. It was decided to use a 50% overexpose from the lowest energy, or 3 seconds, for the remainder of the investigation.

Characterization of Lateral Undercut: The following parameters were held as constant in all patterning and etching: Ø Thermally grown oxide wafers (~5000 A) Ø A one minute 90oC “singe” in contact on hotplate. Ø AZ1512 positive resist Ø Program #1 of the coater was the only coat program used Ø AZ300 MIF developer, used in immersion-develop mode. Ø A 45-second develop, done at room ambient. Ø Stagnant BOE etching, done at room ambient. Ø Mask: EE527 Contact

The following were varied for the purposes of evaluating the extent of undercut in this investigation: Ø HMDS single coat prior to resist Ø Postbake time at 90oC. Ø Etchant concetration (6:1 vs 10:1 BOE)

The experimental outline is as follows in Table [2]:

Wafer

Post-bake time

HMDS

Etchant

1 2 3 4 5 6 7 8 9 10 11 12

0 0 0 0 2 2 2 2 10 10 10 10

None None Single Single None None Single Single None None Single Single

6:1 10:1 6:1 10:1 6:1 10:1 6:1 10:1 6:1 10:1 6:1 10:1

As can be seen, twelve wafers were required. As only six were available, whole wafers were processed through postbake for their allotted time (0, 2 or 10 minutes) and then cleaved along the flat; one half would be etched as a group in 6:1 BOE and the other half be etched as a group in 10:1 BOE. Etch endpoint was detected by the change from hydrophobic-alignment features to hydrophilic features. The results are below in Table [2]. The best performance (via form true to mask, resist appearance good) was observed under the conditions of HMDS, ten minute postbake and 10:1 etchant. In general, all samples etched in the 10:1 held up much better, which is probably due to the etch time differential --- 8 minutes in 10:1, vs. 30 minutes in 6:1 (not all completed etching, but by that time they had failed).

Wafer Number 1

Experimental Conditions

Comments/Observations

No HMDS, 10 min postbake, 10:1 BOE

3

No HMDS, 2 min postbake, 10:1 BOE

5

No HMDS, No postbake, 10:1 BOE

7

HMDS, 10 min postbake, 10:1 BOE

9

HMDS, 2 minute postbake, 10:1 BOE

11

HMDS, no postbake, 10:1 BOE

2

No HMDS, 10 min postbake, 6:1 BOE No HMDS, 2 min postbake, 6:1 BOE No HMDS, No postbake, 6:1 BOE HMDS, 10 min postbake, 6:1 BOE

One half of the features were not clear; there was adhesion loss observed as “links” between the features. Large features were “crisp” All features had cleared, but the shapes were rounded; squares had rounding of the corners, circles were misshapen. All patterns were clear, all were ringed by rainbows (massive undercutting) that touched the adjacent feature. Looked very good! All vias were wellshaped. Diameter of smallest via: 25.28 microns Good! Smallest via: 23.68 micron; appearance of a slight “lip” --- undercut of ~1micron Fair. Smallest via: 21.39. Slightly misshapen circles. Small undercut: ~ 1micron. Horrendous! Very little of the resist survived! Very little here to measure. Like #2.

4 6 8

10

HMDS, 2 minute postbake, 6:1 BOE

12

HMDS, no postbake, 6:1 BOE

Like #2. Half of the sample had resist adhesion failure (gone!). What was left had undercut equal to the small via diameter (10-12 microns) in all directions. Patterns intact. Undercut equal to the small via diameter (10-12 microns) in all directions. As with #10.

Flow chart of the experiment procedures:

Summary and Conclusions The best process flow for creating vias in oxide was as follows: 1 minute 90 C ÒSingeÓ

10 minute, 90 C contact postbake

Spin Coat: HMDS Prog.1

Spin Coat: AZ1512 Prog. 1

Expose: 3 seconds on Oriel Aligner

Develop: 45 seconds in AZ300 MIF

10:1 BOE Etch, approx. 8 minutes.

The process could be yet improved in the following ways: Ø Temperature control of the etch tank and developer solution. Ex.) During our experiment, the room temp dropped to 58 F. This undoubtedly affected the development rate, as the volume used is small. Ø Coat program should ramp up to speed more quickly; there is a large edge-tocenter variance that should be eliminated. Ø The coat uses a lot of resist; roughly 3-5 mL should be adequate. Dispensing the resist in a dynamic mode may give better results. Ø Better control of the exhaust levels in the spin bowl.

You’re only asking for trouble if you… Ø Blow bubbles in the resist bottle with the pipette. The bulb of the pipette should be vacated of air prior to insertion into the resist bottle. At present there is no means to filter out all the tiny little air bubbles introduced --- and these will be a source of pinholes in the subsequent etch. Ø Leave your coated or patterned wafer sitting out for extended periods prior to etching. Water absorption will eventually cause the resist to lose adhesion. Try to pattern and etch in the same day. Ø Rely on a single acetone strip as a rework clean. Xriaxong and I soaked wafers that never received postbake for at least 30 minutes in acetone --- and this wasn’t enough to remove all patterns. Some received a “blanket” exposure in the aligner and develop to remove the pattern and this did an adequate job. Ø Stress the resist film --- no thermal shocks, no bakes above 120 C (N2 evolved in decomposing resist becomes entrapped in the film, imposing stresses).

Ø Use photoresist that has been open for long otherwise it will affect the viscosity of the photoresist and shorten the exposure time and the development time.

References [1]

“Dissolution Kinetics of Crystalline and Amorphous Silica in Hydrofluoric –Hydrochloric Acid Mixtures”, Da-Tung Liang and Dennis W. Readey, J. Am. Ceram. Soc., 70 [8], 570-577, 1987.

[2]

“Determination of the Etching Kinetics for the Hydrofluoric Acid/Silicon Dioxide System”, David J. Monk, David S. Soane and Roger T. Howe, J. Electrochem. Soc., Vol. 140, No. 8, August 1993.

[3]

“Introduction to Semiconductor Lithography”, Mark A. Wirzbicki, Shipley Workshop, Sept. 20, 1995

[4]

“Semiconductor Lithography; Principles, Practices and Materials”, Wayne M. Moreau, Plenum Press NY 1988. (An excellent book!)