|
The laboratories are the heart and soul of Engineering 168. Here you will be stationed week after week, patiently building the device that has defined the late 20th century: an integrated circuit. The steps are relatively simple, and modern manufacturing facilities have been developed that automate the entire process. Unfortunately, Brown doesn't have the prerequisite $18-20 million to buy a new stepper from
Nikon or etcher from Applied Materials.
Instead, the success or failure of building a chip in this class rests entirely in your hands, literally. You will focus and align the masks for lithography. You will load a 950 degree furnace with delicately balanced wafers. You will etch through silicon and metal using hydrochloric acid and sulfuric acid, a drop of which on your skin will send you to the hospital for a couple days. Nevertheless, if everything goes well, at the end of the semester you'll have a little piece of silicon with working transistors to call your own. Who knows, maybe you will be inspired and start a company like
Intel or IBM.
This guide attempts to explain the steps, as well as highlighting the problems and pitfalls that befell our class (the 1997 spring semester). May they help you avoid making the same mistakes as we did! Good luck and happy fabbing!
There are many resources on the net that have to do with fabrication. Here are a couple useful, just in case you have time (and knowing Prof.
B, you almost definitely won't. Well, maybe over summer break...).
Lab 1
-
Time to begin. Always suit up fully before entering the lab! Your skin is one of the dirtiest, oiliest contaminants imaginable, and any dermis contact with an active area of the chip will destroy it. You should cover your dirty self as much as possible: this means gloves, bunny suits, slippers, hair covers, etc. Finally, sneezing on the wafers is bad.
- Clean the wafer, using RCA clean. This is done to eliminate as many impurities as possible from the surface of the wafer. Strong acids are used to etch off all unwanted material. The major hazard in this step is allowing any acid or water to dry on the wafer - if this occurs a non-removable stain may damage any devices that it covers. Thus make sure that a wet wafer is never exposed to the air for an extended period of time. Move it briskly from acid to rinse and back again.
The rinsing is always done in deionized (DI) water, which is extremely
pure. After the process is complete, use the nitrogen gun to dry off the wafer as quickly as possible, concentrating effort on the front and center of the wafer (where the devices are). Again, don't let any water dry on the wafer; push it to the edges and off the wafer. Don't turn off the
flow till the wafer is completely dry, front and back - I can't emphasize how important it is not to let any water or acids dry on wafer. Less importantly, some of these acids are extremely toxic - please use extreme caution when handling them and avoid any contact with the acids. Wear the heaviest latex gloves you feel comfortable in. Prof.
B claims that if you get hydroflouric acid on your skin, you won't feel anything for half an hour, but then it will get to the bone and start to burn with intense pain. If you get ANY questionable substance on clothes, skin, etc., rinse with water for a full five minutes and then make a quick, hopefully short trip to Health Services.
a) 2:1 sulfuric:peroxide mixed (Pirhana etch) 5 min, then DI water rinse
b) 1:1:4 ammonium hydroxide:peroxide:DI at 70 C 5 min, then DI water rinse
c) 5% hydroflouric:95% DI 30 seconds, then water rinse
d) 1:1:5 hydrochloric:peroxide:DI at 70 C 5 min, then water rinse
e) blow dry like crazy
- Grow pad oxide, using wet oxide deposition. When loading the furnace (FIGURE 1), take your time. The huge heat-resistant gloves allow you to get close to the furnace entrance, but they remove all coordination. Make sure you have a good solid grip on the wafer boat; if you are the least bit unsure, step back and try again. The heat can hit your hands all at once, and dropping the wafers is by
far the worst move possible - if you feel nervous, take a couple minutes to get ready. On a counter a safe distance from the furnace, put the wafer boat (see FIGURE 2) into the glass elephant (see FIGURE 3). When you are satisfied that it is balanced correctly, put on the heavy gloves. Carry the elephant over to the furnace, and with the glass rod push the wafer boat out of the elephant into the furnace very carefully. Make sure the boat clears the lip between the glass elephant and the furnace, thereby entering the chamber cleanly and straight. Use extreme caution, and whatever you do, don't panic. If the wafer boat flips over inside furnace, the wafers will probably have to be scrapped.
a) grow wet oxide for 15 minutes at 920 degrees
b) anneal the oxide for 10 minutes at 920 degrees (no oxygen flowing)
 
- After the wafers cool (allow a good 10-15 minutes for this, hot wafers are much more fragile than room temperature ones), measure the oxide thickness with the ellipsometer. The instructions for how to use this tool should be near it.
a) our index was 1.449 and thickness was 676.5 angstroms
- Deposit the sacrificial nitride layer. This can be done in the Plasmatherm or in the LPCVD (low-pressure chemical vapor deposition) furnace. We tried both, as the first run with the Plasmatherm did not produce acceptable results. You may have better luck and not need to use both. First we'll discuss the
Plasmatherm.
5A. The Plasmatherm is one of our facility's better tools, so use it with affection. When it begins to pump down, it is necessary to push down the lid of the chamber when the readout reads around 7.2. This gives the system the help it needs to ensure a
vacuum. After you do this for a second, the pressure does the rest of the work as it goes to high-vac. Enter the correct recipe, and the machine does the rest of the work. One caveat: make sure to open the correct gas tanks, such as ammonia, silane, or whatever else the specific process requires. And if you are using silane, do a nitrogen purge to remove all silane (a gas which reacts with oxygen to form sand) from the gas lines!
a) the first recipe was 168SI3N4, though this might be different
b) 10 minutes nitride deposition
Unfortunately, the layer of silicon nitride laid down by the Plasmatherm proved quite porous; that is, instead of effectively blocking the oxide layer deposited later, the oxide diffused right through, rendering the step useless and the wafers ruined.
5B. This step can also be performed with LPCVD. As discussed above, when we first performed the step with the Plasmatherm, we got very disappointing results. So we tried a second batch of wafers using LPCVD deposition in the furnace, and received much better results. Again, load the furnace in a way similar to Step 2's method, very very carefully.
a) Deposit Si3N4 in LPCVD furnace for 45 minutes
b) Upon removal, wafers should be a rich blue color.
This time, upon measuring the wafers with the ellipsometer it became apparent a good layer of silicon nitride had been deposited. The index was around 2 (indicating high quality). A 753 A layer had been deposited, indicating a 753 A / 45 min = 17 A/min deposition rate. And after some tests, we determined the etch rate to be around 400 A/min with the
CF4/O2 recipe in the Plasmatherm.
Lab 2
-
Lithography of active areas. Litho is the heart and soul of the fabrication process, transferring the pattern from the mask to the chip. For this step, the key is the mask. Use extreme care when handling the mask, as they are worth more than their weight in gold (I think, Brown contracted out to IBM to build them for about $3000). The same masks should be
usable for many years. Make sure the vacuum is turned on; this seals the mask to its holder in the mask aligner.
a) Spin on 1813 photoresist. Use the correct wafer chuck, and make sure there is a good seal. If you can hear any hissing, the seal is probably bad. Lift off the rubber o-ring and replace, with luck the seal will work then. Check the light under the table. If it is on, there is a vacuum and you can go ahead and spin on the resist. DO NOT use a smaller chuck - we tried this and the wafer fractured as soon as the machine was turned on. There was simply too much torque for our cheap wafers. NOTE: if the photoresist does not adhere well to the wafer due to oxide, it may be necessary to use an adhesion-promoting agent called HMDS. This will be spun on the wafer first, then the 1813. After spinning on the resist, bake it on for 2.5 minutes at 100 degrees
on the hot plate.
b) Expose the wafer using the active area light-field mask (see FIGURE 4 and make sure you understand the difference between light-field and dark-field masks. Basically, light field masks let light through everywhere other than the active areas, while dark field masks only expose the active areas). Align one field in the microscope, then check other fields to make sure it's aligned diagonally - it probably will not be the first time. This step takes time and patience, be willing to sit there for 15-20 minutes to get the hang of it. Don't get frustrated, you can align for a long time without getting anywhere and then all of a sudden it just slips into place. You'll get the hang of it. After correct alignment, expose the wafer. Watch the light to make sure the ultraviolet light turns on, but once you verify correct operation look away. NOTE: occasionally the wafer will stick to the mask and thus trying to align will be ineffectual. If this happens, taps the mask very gently and the wafer may fall back to its correct position. If not, remove the mask and carefully pry loose the wafer. Be exceedingly careful not to scratch the mask - if it gets scratched every wafer it patterns for the rest of its lifetime will have the same error.
c) Develop using the MF312 developer. 50-55 seconds worked best for us. Be careful not to overdevelop - if this happens you'll get to do the whole litho process over again! After develop and drying, take a look at the wafer under the microscope. The fields should look clearly defined and sharp. If the edges and corners are rounded, it probably was overdeveloped. If this occurred, see part (d); if not, go on to step 2.
d) Although frustrating, over/underdeveloping is not a disaster. Using acetone, you can strip all the photoresist off the wafer. Make sure you don't use water anywhere around the acetone - the two are not mixable, and should always be kept separate. I guarantee you'll forget this once duing the semester and have an uncleanable dirty beaker as the result. Always have a separate beaker that is acetone-specific, and never try to wash this out with water. After cleaning off the bad photoresist, repeat steps (a)-(c), though don't develop for quite as long.
e) To clean acetone from the beakers, rinse with a more acetone from the squeeze bottle. Then rinse with isopropyl alcohol and dry the beaker well. Again, whatever you do, don't use water!
- Etch the nitride layer in the Plasmatherm. We etched for 6 minutes.
- At this point, all the photoresist should be off the wafer. To make 100% sure the PR is all gone, dip the wafer in acetone. Acetone is terrific for getting rid of photoresist. Follow with an isopropyl rinse and dry well.
- RCA clean, just like the previous lab.
- Grow field oxide, with wet oxidation again.
a) 1150 degrees C for 60 minutes
b) ellipsometer measurement after wet ox: index 1.445, thickness approximately 6800 A.
Lab 3
-
We had a problem with the Si3N4 etch the first time, as the low temperature that Brown's equipment forced us to work with had not etched enough. Another 6 minutes of nitride etch in the Plasmatherm seemed to fix the problem.
- Etch oxide pads. A hydroflouric acid dip was used here.
a) 6-8 minute HF dip
- Grow gate oxide, using dry oxidation.
a) 1150 degrees C for 20 minutes with dry oxygen.
- Polysilicon deposition via low-pressure chemical vapor deposition (LPCVD). We had a lot of trouble with this step, never verifying exactly how much poly was deposited. Pumping down the chamber took a long time due to the old equipment - be prepared to wait a good 20-30 minutes.
a) Again, carefully load the wafers into the elephant and then into the LPCVD furnace, heated to 650 C.
b) Pump down the furnace to around 300 mT. The lower the better, but you probably don't want to wait that long.
c) Set the gas flows to the following settings: SiH4 = 25 sccm; N2 = 25 sccm.
d) Bake for 20 minutes.
After performing some tests, we determined the etch rate of polysilicon in CBrF3/CF4 to be around 130-140 A/min, and the deposition rate of poly in the furnace at 650 degrees to be around 155 A/min.
Lab 4
-
On to the litho of the polysilicon layer. Follow the same steps as the first litho. Again, this is a relatively easy alignment as the mask is a light-field.
a) spin 1813 photoresist
b) expose with polysilicon light-field mask
c) develop. 30 seconds seemed about right - 35 seconds slightly overexposed the wafer.
- Etch the oxide and polysilicon in the Plasmatherm. We want to take off about 3800 A, so at the rate of 130 A/min (calculated in tests above) we need to etch for about 30 minutes in CBrF3.
a) Gently place the wafers in the Plasmatherm, then use recipe 168CBRF3.
Our wafers moved slightly in the Plasmatherm during processing, resulting in an uneven etch. Nevertheless, in our best wafer, we had a clearly defined 3400 A high gate stack. From a visual standpoint, our wafers had a bluish color towards the edge, a green field near the center, and pink gates. Be careful not to overetch in the buffered oxide etch!
Lab 5
-
A quick lab, another lithography to define the p-channel active areas. Same procedure as the first two lithographies except that this mask is a dark-field mask (again, see FIGURE 5 and verify you are using the correct mask!). It is slightly more difficult to align, as the whole mask is dark except for the openings. Use the large openings to align first, then use the smaller features.
a) spin 1813 photoresist.
b) expose using the n-channel active area mask.
c) develop.
- At this point, the wafer is sent California to be implanted with boron. Brown's meager resources mean we don't have an implanter, so you get a two week respite. This will probably be timed to coincide with spring break.
a) the parameters we used: 20 keV implant at a doping level of 7*1014 atoms/cm3, with a 7 degree tilt.
Lab 6
-
The wafers should be happy after their ordeal in California, a little tan perhaps. To help them get re-acclimated to Providence's weather, they will be annealed to activate the boron implants.
a) 1000 degree C anneal, 15 minutes in inert N2
- Grow oxide to cover gate. Use dry oxidation.
a) 1000 degree C, 15 minute dry oxidation
- Cover the entire wafer with oxide, using Plasmatherm oxide deposition.
a) our process: 220 degrees C, 5 min
b) ellipsometer measurement: 1154 A
Lab 7
-
Almost there! This lab is to define the interconnects. Lithography is performed first, again with a dark-field mask. This is an absolutely critical layer: misalign slightly here, and you'll doom any smaller devices.
a) spin 1813 PR. You should be getting the hang of this by now!
b) expose.
c) develop. Again, always double check your alignment under the microscope after the wafer has been developed and dried to make sure it was correctly aligned and not over/underdeveloped. In our lab a perfect alignment looked horrendous on double checking, as it apparently slid a bit after alignment during exposure.
- Etch contact holes, using buffered oxide etch.
a) Use the buffered oxide solution. This is a strong etching solution, and in fact broke one wafer. But the wafer had already been weakened due to an unfortunate collision with the floor (it was dropped). All wafers that hadn't been banged too much survived. According to Prof. Z., the wafers he used were the cheapest he could find from Poland and much thinner (and thus more fragile) than the specifications called for.
- Deposit blank metal - aluminum. Use the evaporator to do this. The evaporator uses a rather simple technique, see FIGURE 5. Basically, Al is heated in a relatively good vacuum, around 10^-7 torr. The wafers are placed on a metal sheet which is set on top of the container, and the evaporating aluminum inundates each wafer with an even layer of metal.
a) Prof. Z. set up the tool for this step. The goal was approximately 2000 A.
Lab 8
-
Last fabrication step! It's another lithography lab, but a light-field mask. In this lab we define the metal contact pads.
a) spin on PR. We used CD-30 developer for this step rather than the standard 1813 used in the other steps because 1813 attacks aluminum - which we will be using as an etchant for this step.
b) expose for 1 minute. Seemed to work well.
c) develop. This took some experimentation. We first tried MF312, but it was too slow. We switched to an aluminum etchant, which was too slow at well. Finally Prof. Z. hit on the right solution: heat the aluminum etchant. At 50 degrees C, the etchant worked at a very good pace, etching off the blanket of aluminum (which had been evaporated on the wafer in Lab 8)in approximately 1:05 (m:s).
d) strip the remaining PR with acetone.
- All done. At this point, the devices are complete. Yee haw!
Lab 9
-
To make the wafers testable, they must be cut and placed on glass slides. Also, a contact must be made to the bottom of the silicon to act as a reference ground.
a) cut the wafers. Prof. Z. may do this; if you have to, use the scriber and apply slight pressure to the edge of the wafer. With luck, the wafer will split in a clean, straight line.
b) scratch oxide and other junk off the back. Use the scriber to carefully scratch at the back until you are positive a good amount of silicon is exposed. DO NOT scratch too hard - these wafers are extremely prone to breaking!
c) add hot indium to two/three sites on the bottom of the wafer (that you just scratched), and verify that they are actually contacting the silicon. Check this by measuring the resistance between two sites. They should range anywhere from .5 kohms to 5 kohms.
d) glue the chip to the glass slide using the indium. Liberally apply indium; it's cheap, and a good connection is critical to guarantee testability. Leave some indium on the side of the chip to use as a contact.
Lab 10
-
There's always a caveat, of course: the wafers need to be tested. Use the Hewlett-Packard semiconductor tester; Prof. Z will provide information about operation this machine.
a) Test all devices that look good. The small transistors (2-4 um) will probably not work, but the larger ones should. However, they may not work well: ours didn't turn on till about 20 V was applied to the gate. Obviously this is much higher than the desired 5 V. Try to guess why devices didn't work, or why they didn't work well. In our case, it was probably because the polysilicon deposition was uneven at best.
- Test the capacitance of test sites.
a) There is a capacitance measuring device that Prof. Z. will demonstrate.
Take a chip home and hang the tiny thing on the wall. Heck, you built it with your own two cents. Intel makes $30 ($US1996) billion a year building the devices - that's a good couple years of Brown's tuition! Enjoy telling your grandkids about the experience of Engine 168.
|
