You're looking at companies like Citibricks doing wafer scale, but again, it's larger, but it's still programmable, but it's still a model has to fit on the wafer, right? And then there is a number of other ways to go about it, whether it's digital in memory or others, right? So how do you see your solution kind of fitting in with those types or even generic AI kind of standard solutions? So I would

say...

To bring more bandwidth to compute, there have generally been two approach.

One is through packaging.

So HP high bandwidth memory is a packaging solution that packets a lot of DRAM chips and use SERDES to bring data in.

And there are 3D IC ways of attaching additional SRAMs or DRAM onto the compute.

So packaging is definitely a key element to that.

Yeah, happy