quantized
Member level 2
- Joined
- Jul 6, 2012
- Messages
- 51
- Helped
- 1
- Reputation
- 2
- Reaction score
- 1
- Trophy points
- 1,288
- Activity points
- 1,887
As we all know, the cost of a photomask set has gotten pretty outrageous these days; in the millions or tens of millions of dollars for advanced processes.
The explanation I hear every time I ask about this is that e-beam writing machines are incredibly expensive (I believe this) and the time to write a mask tends to scale with the number of features on the mask (which increases with each generation) rather than the size of the reticle (which doesn't change much). So that machine has to pay for itself, and so if it takes 0.0001% of the e-beam writer's lifetime to write a mask, the mask had better cost at least 0.0001% as much as the e-beam writer.
If this is true, shouldn't it be possible to buy a maskset on which you've only written a pattern into a tiny corner (say, 5% of the mask area) and the rest is left blank?
This way you ought to be able to try out a design with what amounts to a "one project MPW". It ought to cost ~5% of the cost of a mask and the cost of one lot of wafers. This is much, much, much, much less than a full set of masks on every process I've ever worked with.
So, what's wrong with this explanation? Are the fabs hiding other account-handling costs in the mask fee? Surely it isn't the cost of the raw material of which the mask is made (quartz, cheap stuff). I suppose you'd need fill patterns on some layers, but there are ways around that if it truly is the only issue.
Being able to test out tiny parts of a design without having to wait for the next MPW (and the time required for all-layers-enabled MPW runs) would radically change IC design. It would also reduce the support load on the process engineers -- a lot of questions I ask them I could answer myself by simply creating a test circuit. In fact, if you managed to get the minimum run cost and turn-around time low enough the fab could offload a lot of its PDK development+maintenance effort to third parties.
(circuits guy who does not know all that much about fabrication)
The explanation I hear every time I ask about this is that e-beam writing machines are incredibly expensive (I believe this) and the time to write a mask tends to scale with the number of features on the mask (which increases with each generation) rather than the size of the reticle (which doesn't change much). So that machine has to pay for itself, and so if it takes 0.0001% of the e-beam writer's lifetime to write a mask, the mask had better cost at least 0.0001% as much as the e-beam writer.
If this is true, shouldn't it be possible to buy a maskset on which you've only written a pattern into a tiny corner (say, 5% of the mask area) and the rest is left blank?
This way you ought to be able to try out a design with what amounts to a "one project MPW". It ought to cost ~5% of the cost of a mask and the cost of one lot of wafers. This is much, much, much, much less than a full set of masks on every process I've ever worked with.
So, what's wrong with this explanation? Are the fabs hiding other account-handling costs in the mask fee? Surely it isn't the cost of the raw material of which the mask is made (quartz, cheap stuff). I suppose you'd need fill patterns on some layers, but there are ways around that if it truly is the only issue.
Being able to test out tiny parts of a design without having to wait for the next MPW (and the time required for all-layers-enabled MPW runs) would radically change IC design. It would also reduce the support load on the process engineers -- a lot of questions I ask them I could answer myself by simply creating a test circuit. In fact, if you managed to get the minimum run cost and turn-around time low enough the fab could offload a lot of its PDK development+maintenance effort to third parties.
(circuits guy who does not know all that much about fabrication)