MgF2 Burst Pressure Analysis

ByGreg Madejski

Key Ideas

  • I took a wafer with ~400 chips of varying square window sizes, ran my MgF2 process, and recorded the pressures it took to burst them.
  • Overall yield was about 20%, most of them smaller windows. I was expecting ~50% based on my previous work. Variation in the etch step destroyed the yield.
  • MgF2 nanomembranes are weaker than pnc-Si,  ~25% as strong.
  • MgF2 nanomembranes don’t follow the same exponential dependence observed for pnc-Si and NPN materials.
  • Overall wafer yield summarized in a GIF (colored = burst membrane):

output_034sS2

 

Screen Shot 2015-01-30 at 4.07.25 PM
My handling was not terrible (Dicing, Before Burst Pressure Measurement yields above 85%). The Evaporation and Annealing went fine (historically 50-80%), but the etch process (historically 50-80% with updated etch recipe) went poorly.

 

Initial Wafer

 

Burst Pressure Wafer true orientation
Mask Orientation. The wafer is a 6 slice pie, with varying window sizes (0.1, 0.3, 0.5, 0.7, 0.9, 1.1) in each slice, 66 chips each.

 

IMG_2653
Starting Wafer 1163, NPN, 23% porosity, 38.5 nm average pore diameter.

 

 

Deposition

Using the Kurt Lesker Sputter/Evaporator, I deposited 50 nm of MgF2 on my wafer. Base pressure was 5e-5 torr, and I used the 6 KV oxide recipe, with platen rotation on, at room temperature. I controlled the current manually, varying the current between 0.5-2.0 mA to achieve stable deposition rates. I tried to maintain a rate between 0.1-0.2 nm/s, but I occasionally had spikes up to 0.45 nm/s, unusual in this recipe. The coloration of the film was a deep purple, as I have noted before for this film thickness/substrate. No broken windows were observed at this point.

 

Screen Shot 2015-01-30 at 4.01.57 PM
No broken windows after MgF2 evaporation.

 

Discoloration around edges where the shadow mask in the evaporator did not cover the 6" wafer. With platen rotation on, it homogenized the additional film thickness as the crucible reached a stable temperature.
Discoloration around edges where the shadow mask in the evaporator did not cover the 6″ wafer. With platen rotation on, it homogenized the additional film thickness as the crucible reached a stable temperature.

IMG_2657

 

Cleaving

Sectioning the wafer into smaller pieces to fit into the 3″ tube furnace was nervewracking, but overall, I didn’t destroy that many chips in the process of doing so. The major crack I identified was in the bottom left quadrant, and my first broken chips came from there.

Screen Shot 2015-01-30 at 4.01.42 PM

 

 

Annealing

I ran a 2 hr recipe to anneal the MgF2, in 15 sccms of Ar as an ambient. Insertion (t=0) occurred at 200C or below, and then ramped up to 600C.  Removal (T=2.5 hrs) occurred at 450C or below. I lost the locations of  a single large block of chips when it clipped the edge of the furnace. User meeting attendants know my consternation.

 

Screen Shot 2015-01-30 at 4.01.29 PM
Chips in the Red region were dropped, so the precise location is unknown. The locations are accurate based on the window size of the chips. Lower yields are present around the edge, as expected from the heavily stressed evaporation process (blue ring).
IMG_2664
Where large regions of chips were preserved, I tried to keep them together during the anneal (this is more like how a whole wafer would be annealed at once). Single chips were annealed on a broken wafer I fit into the 3″ tube furnace.

 

Etching

I flipped the chips over and removed the silicon nitride from the backs of the deposited films. My etch yields were terrible.  I use a mixture of CHF3 and 7-10% O2 at 75 mTorr at 100 W. This provides an etch rate of 50-60 nm/min. I cleaned and seasoned the chamber each day with an Ar etch for 10 min, followed by a 60s etch using the above recipe, tuning in an acceptable run. I lost a great number of chips in each box with the 70 s etches, and the power spikes. To try and prevent time for the power supply to spike, I lowered the etch time to the absolute minimum to get rid of a nitride film. I was unable to find what was causing my plasma to spike wildly occasionally, and not other times. Box 7 had zero yield. At this point, I had not quantified a maximum power associated with broken chips. If 120W is a practical limit, I am operating fairly close to that power ceiling. I will have to see if I can make a recipe that is lower power, stable, and maintains the low etch pressure and gas ratios. The tradeoff would be potentially increased polymer formation and of course longer etch times.

Box Date Pressure (mTorr) Average Power (W) Max Power (W) Time (s)
1 1/21/15 79 98 101 70
2 1/22/15 82 110 115 70
3.1 1/22/15 79 98 170 70
3.2 1/22/15 80 97 115 70
3.3 1/22/15 78 96 99 55
4 1/23/15 78 96 101 55
5 1/23/15 81 98 112 55
6 1/23/15 78 98 105 55
7 1/23/15 82 101 124 55

 

Screen Shot 2015-01-30 at 4.07.25 PM

Screen Shot 2015-01-30 at 4.01.13 PM
The numbers shown here are the box the chip was in, followed by the etch run it was broken in (ie 6 vs 3.1). I etched each box separately, except #3 which I split into 3 runs to see if the etcher was stable. Box 7 had 0 yield.
Position in the etch chamber
Position in the etch chamber (post-etch)

 

 

Burst Pressures

I then recorded burst pressures using the same setup as Sarah used in her work. Instead of screwing the chips into the clamp, I used a hard clamp to make sure the same pressure was applied to the chip everytime.

Burst Pressure Wafer

Screen Shot 2015-01-30 at 4.00.52 PM
Heat Map shows a definite dependence on window size. The zero values are chips I haven’t measured yet.

 

IMG_2671

I 3d printed hard c-clamps to ensure the same pressure was applied to each chip.
I 3d printed hard c-clamps to ensure the same pressure was applied to each chip.

 

Analysis

Screen Shot 2015-01-30 at 4.07.25 PM

Window Size (mm) Average (PSI) Std Dev (PSI) Count Coef. Of Variation Relative Yield
0.10 15.42 16.20 36 1.05 55%
0.30 7.78 4.22 7 0.54 11%
0.50 1.43 1.55 5 1.09 8%
0.70 2.99 0.91 15 0.30 23%
0.90 1.50 0.67 6 0.45 9%
1.10 1.76 0.94 12 0.54 18%

Burst Pressure Raw Scatter

The graph includes everything that I measured, including the chips where the thickness of the film is unknown, but almost certainly occluded (blue ring). In order to get a more precise measurement of my process, I removed data points that were within 2 chips of the edge of the wafer.  About half of the chips measured are thrown out. The coefficient of variation is reduced for a few window sizes, indicating a tighter distribution.

 

Window Size (mm) Average (PSI) Std Dev (PSI) Count Coef. Of Var. Relative Yield
0.1 6.40 2.05 18 0.32 27%
0.3 5.31 3.70 4 0.70 6%
0.5 2.04 1.84 3 0.90 5%
0.7 3.16 0.93 11 0.29 17%
0.9 1.50 0.67 6 0.45 9%
1.1 1.59 0.62 7 0.39 11%

Burst Pressure Edited Scatter

This whole graph would fit in the bottom portion of the first graph.

Burst Pressure Edited Column Trend
Error bars are 1 standard Deviation

 

Compared to pnc-Si and NPN, the MgF2 is much weaker, only a quarter as strong as pnc-Si at the smallest window size. The power regression is broken here as well, with the exponent only being 0.6 compared to 1.2.

All Materials RAW All Materials Chart

 

If we run another burst pressure wafer, we might get cleaner data from higher n. First I need to solve the problem of the etcher.  Overall, this study reinforces what we’ve practically known for a while: my MgF2 nanomembranes are delicate. We have been able to grow things on them in vitro, but bubbles will wreak even more havoc on them compared to silicon materials.

Similar Posts

Leave a Reply