Presumably they make these things by the gazillion on a thin silicon wafer, and then have to somehow cut them up into individual chips 0.05 mm on a side. This is beyond the capability of even my finest hacksaw, and I’d think lasers would cook the circuit, while water jets would send the cut powder flying across the cleanroom. So how do they cut them?
With a nickel diamond grit blade, according to a paper I found, which is in IEEE XPlore, so I won’t link. Here a page for a company that does it. I know nothing about them.
There is relatively a lot of space between dies on a wafer, but a quick Google revealed a lot of papers about controlling the process and limiting damage.
I’d bet they have some sort of vaccuum running during the sawing to pick up the dust. I’ve seen yield numbers at various stages, and I’ve never seen damage during dicing as a problem.
They may have a layer of something soluble holding all the devices, then dissolve that layer away. By “soluble”, I mean soluble in something they have in IC manufactories. It could be Aqua Regia, Piranha Etch, HF…
No, the dies are an inherent part of the wafer, produced by stepping the mask over the different die locations. You might be thinking of process steps where they deposit something then etch away all the parts not wanted. That isn’t the process for splitting the dice into wafers.
You say that there’s relatively a lot of space between dies on a wafer, but even with that, they still end up having to make 4 cuts around each tiny circuit, and the last cut will be on a strip thats already been trimmed to 0.05 mm along one axis. Won’t it? They must do something to control the exact position of these tiny pieces. How thick are the nickel/diamond blades?
click on facilities at the link I provided, and look at the specs on the saws they have. The K&S 982-10 has a cut placement accuracy of 4 microns, with 3 sigma variance. The resolution is 0.8 microns, and the accuracy is 2 microns.
It also seems they use a wash to take care of the residue, and they saw in a clean room, of course. Around the edge of a die are the I/O pads, which are much bigger than the usual logic cells, having to drive and receive signals off chip. If you look at a die photo, you’ll see there is plenty of leeway. Or plenty when you think in the terms you have to think in for deep submicron.
I don’t know how they build these saws, but I bet they’re plenty expensive.
Those specs are amazing, down in ruling engine range for Diffraction gratings. There must be a big patent literature on the design of these saws. It’s time to do some digging. Thanks for the link,Voyager.
Looks like the OP still has not been answered. In my experience (I work in this industry, though not for Hitachi), the wafer is covered with a protective overcoat of oxide and/or nitride, then it’s thinned down substantially by grinding the backside, then it’s placed on a layer of plastic adhesive tape. The saw then, very precisely, cuts through the silicon wafer leaving the tape intact with the cut dice adhering. Later, each good die is picked off and placed into a package with an automated tool. It’s possible that Hitachi has developed improvements or refinements for the ultrasmall die-size you mention.
>No, the dies are an inherent part of the wafer, produced by stepping the mask over the different die locations. You might be thinking of process steps where they deposit something then etch away all the parts not wanted. That isn’t the process for splitting the dice into wafers.
No, no, that’s not the point. Normally you split the wafer into dice (not the other way around) by sawing it, but then normally you continue to handle each one separately to test it, bond leads onto its edge, place it into a package, etc etc. I think the whole idea here is that these dice are so small they’d look like specks. You certainly couldn’t tell they were squares, for instance. And an RFID device doesn’t need any leads at all (at least not necessarily) and may not need to be installed in its own package. So, therefore, in this case they might use dissolving as a method, even though it normally wouldn’t be. When the question is, “How do they do this strange new thing?”, you can’t eliminate a possible answer just because it’s not the standard way they’ve always done something else.