REGISTER | LOGIN
Breaking News
Blog

Time to Look For Low-Cost DRAM Alternatives

NO RATINGS
View Comments: Newest First | Oldest First | Threaded View
Page 1 / 3   >   >>
Sang-Yun
User Rank
Blogger
Re: 3D cost
Sang-Yun   12/4/2017 2:24:58 AM
NO RATINGS
AM, as you know, litho is just one of many cost factors. Even though DP is expensive in general, overall process is a just short loop and, therefore, it has high wafer throughput. Unlike other DP for logic, the DP for 3D Super-NOR is relatively much simple for process control. Sorry, I cannot be more speicifc in public. 

double-o-nothing
User Rank
Author
Re: 3D cost
double-o-nothing   12/4/2017 2:12:32 AM
NO RATINGS
1 saves
Sounds like 5x the 20nm double patterning. Has the multiple patterning gotten cheaper, over time?

Sang-Yun
User Rank
Blogger
Re: 3D cost
Sang-Yun   12/3/2017 5:00:43 PM
NO RATINGS
5-layer using 20nm feature size. Die size is about 100mm2. Target cost is about 3 cents/GB and target price is about 6 cents/GB.

resistion
User Rank
Author
3D cost
resistion   12/2/2017 8:23:44 PM
NO RATINGS
How does 3D cost scale with number of layers? Seems would approach proportionality for a high enough number of layers.

resistion
User Rank
Author
3D cost
resistion   12/2/2017 8:15:06 PM
NO RATINGS
How does 3D cost scale with the number of layers? Seems to approach proportionality with a high number of layers.

Sang-Yun
User Rank
Blogger
Re: Where will these cheaper alternatives to DRAM be made?
Sang-Yun   11/26/2017 10:00:45 PM
NO RATINGS
Realjjj, thank you for sharing your inputs. I agree with you. Cost-per-bit is the rule of thumb for memory products. So, cost scaling for the next 20 years should be considered and practcally implemented by products. And these produts should be introduced to market in short time. That's what I working on it. 

realjjj
User Rank
CEO
Re: Where will these cheaper alternatives to DRAM be made?
realjjj   11/26/2017 9:35:59 PM
NO RATINGS
Sure but how do you define a winner in memory? For me the most important metric is scaling, at least nowadays -horizontal, vertical, bits per cell. For a new solution, it's not sufficient for cost and perf to be good enough today, it has to scale very well for a decade or two. That's why I am skeptical about XPoint or embedded MRAM but hopeful when it comes to WD's 3D ReRAM.

If the main issue is scaling, something new is an opportunity to not only keep up with Moore's law but beat it, 3D can in theory enable that. If horizontal goes with Moore's, adding layers and bits per cell could come at a low cost and enable much better than Moore's in an ideal scenario. So if you have X costs today, where are you in 10 years, at X/10 , X/32 or X/1000- 1000 times lower costs in 10 years would imply a 50% cost reduction per year.

Sang-Yun
User Rank
Blogger
Re: another basic spec
Sang-Yun   11/24/2017 11:25:22 PM
NO RATINGS
Tanj, regarding power consumption, NOR has about 1/8th of DRAM power comsumption in general. As you pointed out, overall power consumption of 3D Super-NVDIMM should be carefully reviewed in order to fit into existing systems

Sang-Yun
User Rank
Blogger
Re: basic specs for memory replacement
Sang-Yun   11/24/2017 11:22:02 PM
NO RATINGS
Tanj, you are right, 1;1,000 ratio is too extreme. In Figure 3, I mentioned "10GB DRAM + 1TB NOR" which is 1:100 ratio.

Sang-Yun
User Rank
Blogger
Re: basic specs for memory replacement
Sang-Yun   11/24/2017 11:19:06 PM
NO RATINGS
Tanj, regarding the worst case scenario, your assumption is right. "DRAM+NOR" is not the same as "All DRAM" at the same memory density. Though, considering cost vs. performance, 3D Super-NVDIMM (i.e. DRAM+NOR) should be a very attractive choice. And, it is non-volatile. So, 3D Super-NVDIMM is the replacement of traditional "DRAM+SSD" memory hierarchy.  

Page 1 / 3   >   >>
Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed