Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

The most amazing thing about this (and another tiny RFID chip that was on HN recently) is not that you can print them on wafers, but that you can cut up the wafers and handle these tiny dies. Imagine you manufactured sugar, but had to manipulate each sugar grain separately.


It thought it was an interesting analogy, so I looked up the size of a typical sugar crystal. It's between 450 and 600 microns. So these chips are 3 to 4 times smaller than that even.


Checks out with the picture in the article.


I was in a project with MAN Roland and the university of Dresden at the same time and the most important thing is roll to roll printing. This actually works well if you do not need an external antenna. What was the holy grrail at the time was to print also the antennas, get a decent coupling and then actually also do item and not only batch level tracking of the packaging you would print. Particular Pharma was really interested in terms of anticounterfeiting at low cost.


I have always wondered how this works (along with wire bonding), especially in an economic way.

Chips being cheap makes sense at the lithography / wafer level because sure, you can stamp out thousands of them at once. But once you need to dice them up, bond wires to them, and package them... how on earth do you do that so efficiently that each chip can be sold for fractions of a cent?


Packaging used to be a huge portion of the industry, in the 80's when pin counts exploded and litho got cheaper it was usually the most expensive part of a chip.

Today, it's cheaper mostly because of flip chips and Wafer Level Chip Scale Packaging (WLCSP). You build the bond pads as a normal litho step, and use a dielectric that's non-wettable between them. Then you can just use a mask to produce a grid of solder balls in the right places, drop the chip on them and put it in an oven. When the solder melts, the chip will self-align on it, so long as it's not too far off. It's uncanny to see it move.


Lots of automation. Dicing is automatic, bonding, testing are automatic. The manual work is mostly just transporting materials.

The bonding machines are crazy. Definitely look it up on YouTube, the machine puts down bond wires super fast.

The other part of it is sheer scale. Once you start making thousands or millions of something, economies of scale drive the costs way down


And for reference there's a section in this BBC film about how it was done in the 1970s, by hand: https://www.bbc.co.uk/programmes/p01z4rrj


Bond wires really are a thing of the past now except in niche applications or legacy chips. Flip chip designs have the die directly interface with a substrate that is then soldered to the board. With the kinds of speedy signals modern chips use, bond wires introduce impedance mismatch that degrades quality. This is related to why you don't see many new designs using pins.


If you can shine lasers at them to etch out a chip trace, then you can shine lasers to carve up the chip. Just use stronger lasers.


Well, these ones are RFID chips, so they don't need to be connected at all - they are read wirelessly.


They'll still need an antenna.


If you want the bare chips and not full assembled labels the usual packaging is uncut wafer and cutting out and handling the individual dies is your problem.


> If you want the bare chips and not full assembled labels the usual packaging is uncut wafer and cutting out and handling the individual dies is your problem.

This does not match my experience, although I imagine it's true in some parts of the industry. I've seen bare dice usually delivered as KGDs (known good dice -- tested at the individual level, not just the wafer level). These used to be shipped in waffle packs, but I've more recently seen blue tape used for delivery, and going straight into pick-and-place.


How big is the saw if you're cutting up 150 micron pieces of wafer?


Sometimes a diamond saw is used but modern processes use lasers to score the die and then the wafer carrier expands to break along the scores.


This is the really interesting Thing!!!! And: how they can have different ROM content (code) for each chip


You significantly underestimate the size of sugar crystals ^^




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: