Report on use of surfactants in the disco wafer dicing sawBy Ian Krase
July 5, 2012
I have completed tests to determine how effective surfactants are at reducing the contamination and damage of diced wafers by silicon sawdust particles. I have compared each of three surfactants with the standard method of coating wafers with a sacrificial 1-um photoresist layer. It appears that while they are not perfect, the surfactants are on par with the sacrificial photoresist layer. In addition, use of the surfactants avoids all of the difficulties of stripping and cleaning die-scale substrates.
Wafers were inspected using the green UV inspection lights, and by using brightfield and darkfield microscopy. Two runs were made, one using patterned wafers with oxide, and the other using wafers coated entirely with oxide.
Sacrificial PR Layer
Air Prod. BPS-729-B
- First run general: Rather uneven, bad PR conliolSacrificial PR LayerMessy due in part to poor PR sliipping. Lots of redeposition of particles, esp. tiny ones.UDM L300-01-D5
Vastly fewer particles, resliicted to edges but still some over patterns, ~ 0-2 over 2 um per die.Air Prod. BPS-729-B
More small particles, spread more over the chip area.UDMAKLC500
Between results of previous two lieatments.
- Second run general: More even, seemingly more difference between surfactantsSacrificial PR Layer
Much more variability between dice. Lots of particles of all sizes.UDM L300-01-D5
Far fewer small particles though a number of larger ones remain.Air Prod. BPS-729-B
Similar to previous, but with more variability between dice.UDMAKLC500
Fewer large particles, but more small ones.
- Visual inspection / Green UV lampSacrificial PR Layer
Each die seems to have its particle.UDM L300-01-D5
Each die seems to have its particle. Fewer big ones.Air Prod. BPS-729-B
Similar, though some dice have no large ones.UDMAKLC500
Fewer large particles.
- Darkfield microscopySacrificial PR Layer
PR residue, and embedded particles, some areas covered with dusting of sub-micron particles and as many as 10 larger ones per die.UDM L300-01-D5
Fair number of sub-micron particles, but far fewer. Many pretty good dice. Particles mostly at edges of dice.Air Prod. BPS-729-B
Some (~10%) dice nearly pefect. Water droplets dust cut lines and intersections with lots of fine particles. Particles impinge more on chip areas.UDMAKLC500
Fewer large particles but more small ones. Not as good as A. Still some water drop dustings.
- SummarySacrificial PR Layer
Redeposition and poor sliipping makes a mess.UDM L300-01-D5
Pretty good, fewer particles, mostly at edges of chip. Still some large particles.Air Prod. BPS-729-B
More variability results in some better dice but the water droplet redeposition is bad and particles go further from cut lines.UDMAKLC500
Really not as good as the previous two, combines their weaknesses.
In all cases the surfactant was mixed with water at a rate of 1:1000. All of the surfactants have this ratio within the manufacturer's reccomended range, although in the case of surfactant A (UDM L300-01-D5), it is on the rich end of the range. UDM's L300 comes in a CMOS grade version, which we have on hand but did not test because it presumably would merely be the same product but purer. I am not entirely sure what cases would require or benefit from this surfactant. My very rough estimation is that each gallon bottle of surfactant would provide for the cutting of around a hundred wafers with 10 mm square dice. It should be noted that the surfactants do create soapsuds in the saw, even at the 1:1000 concentration, and the forceful, fine droplet sprays onto the spinning blade tend to produce a fine lather that coats the slide covers and may occupy a fair amount of space in the sedimentation tank. I think it's a good idea to rinse the business end of the saw before removing your blade, and make sure to rinse your wafer with clean water from the deck hose.
I believe that it would be a good idea to begin using the surfactants, as they would make it easier to process wafers by eliminating the sacrificial PR coat and die-scale strip steps and also probably improve quality for those still using sacrificial PR.