What is harder to make?

[Duke3d87]

I have a question that i have been thinking about (hope this is in the right section)

What is harder to make? DRAM, NAND or logic (i.e CPUs, etc)

The ITRS uses DRAM half pitch for its projections, but if you look at 2012, it calls for a half pitch of 32nm.

Intel has been using the 32nm process for almost two years (logic) and is about to start shipping 22nm devices. TSMC has 28nm parts.

Samsung has been using 3xnm process technology for maybe a year. Elpida/Hynix has been mass producing it for maybe a year or two.

Then comes flash. In early 2010, IMFT announced their 25nm 64 Gb NAND. Recently, they talked about 20nm flash. Toshiba/Sandisk has had 24nm NAND for a while now and have talked about transitioning to 19nm.

So from the process side, what is harder to make? And would the tool sets between a logic and memory fab be significantly different?

 

[RamonGTP]

I'm pretty sure CPU's are the most difficult to make. I think the reason you see CPU's move to a smaller manufacturing process first isn't because it's easier for them to do so, but because of necessity.

 

[Duke3d87]

You'd think that with DRAM it'd be more of a necessity.

In February, Elpida filed for bankruptcy (and has ~$5 billion debt). Hynix and Micron both posted a net loss this past quarter.

 

[RamonGTP]

Not sure what your point Is. I'm talking about a necessity to evolve. CPUs need to get faster and faster while consuming equal or less power.

 

[Duke3d87]

I am not talking about the need to push performance, rather the necessity to keep your company afloat. DRAM isn't a very profitable market right now, so one would argue that your process technology is key.

Yet it seems like it has not kept up with the logic process shrinks and I am wondering if there is something inherently different about DRAM processing that explains the gap. Or on the flip side why NAND seems to be trending slightly ahead of the logic curve.