Friday, December 30, 2016

QLC or 96L 3D NAND... which is better

As we look into the future, one question is:
"What is the best way to scale costs in 3D NAND?"... more layers or more bits per cell?

The assumption is that we will keep adding layers, use string stacking as needed. We also will go to QLC or 4Bits per cell. We will do both. In reality companies have to decide where to focus efforts and jumping to increased layers and QLC at the same time might be difficult to engineer and market.

String stacking adds wafer cost and die size. QLC could control both but will limit write and potentially read performance even lower than TLC. Going from 64L to 96 adds theoretical 50% more bits. Going from TLC to QLC adds theoretical 33% more bits.

Cost estimates based on a 768Gb die in 2018 production are available with scenarios favoring one or the other based on yields and die size estimates. Details are available with all the numbers and yields and cost.... but most scenarios show that QLC will need design changes to prevent reduced performance to stay competitive.

MKW Ventures Consulting, LLC

Tuesday, December 20, 2016

Top 5 NAND and NVM Myths for the New Year

Top 5 NVM and NAND Myths for the New Year

1) NAND prices are dropping and SSDs will be cheaper per GB than HDDs soon.
REALITY: NAND prices were dropping … now flat to going up. HDD prices are dropping. SSDs are still 5-7x more expensive per GB.... and the gap really isn't closing very fast. NAND prices over time will drop 25% per year and HDD prices will drop 10-15%. No crossover coming soon. Whether consumers will take a 256GB SSD over a 1TB HDD will be decided in 2017... place your bets (I have mine placed and can tell you what is happening)

2) NAND SSDs have instant on capability
REALITY: NAND SSDs do boot faster. 50% faster to be exact. But it still takes significant time from power off. Are consumers willing to pay for 20-30 seconds of time per day? NVMe 3DXP and RRAM SSDs will boot even faster. It will take 10 seconds for them vs 15 for a NAND SSD and 30 for an HDD. Boot times

3) 3D Xpoint will replace NAND in next 2 years:
REALITY: 3D Xpoint and the technology behind it are faster and have advantages. But we always knew that new technologies ramp much slower than original commits from the inventor. Sometimes they never ramp. 3D Xpoint is behind Intel’s original schedule (but actually is matched to Micron’s original schedule), so be patient. NAND is lower cost today and tomorrow but RRAM and 3DXP will give it some competition 3 years from now (See my article on RRAM Cost)

4) 3D NAND will cut costs dramatically
REALITY: 3D NAND will reduce costs by 25% per year over planar. Which is exactly the trend for Planar for the last 10 years (FYI: DRAM costs dropped 30% per year). So 3D lets memory cost scale like it always has scaled… assuming companies have a successful, mature 3D NAND Process. TLC 3D Ramp is the big thing in 2017. I have exact bit% and cost numbers as well.

5) Multiple New companies will enter NAND/NVM market
REALITY: New companies will be <3% of the market 3 years from now. NAND is hard, 3D NAND is really hard, getting it cost effective is nearly impossible. Give credit to the existing companies that contribute >99% of the NAND bits, they evolved and do a great job. New companies in China have huge advantage in capital, but until they get technology from a current industry leader, they won’t have a decent technology. If they sign a JV with Micron, Hynix, WD… that would be different (call me about whether that is going to happen)

3D NAND, TLC, More SSDs in 2017… it will be interesting

MKW Ventures Consulting, LLC

Thursday, December 1, 2016

Update on FMS 2016 Predictions Part 1

At Flash Memory Summit (FMS) 2016, I was on a panel making predictions. its been over 3 months... Update part 1

SSD cost/GB drop 25% per year (On track!) 
Prices 25% or more  (Off track for now... but just wait... )
TLC becomes dominant in all SDDs 
Client and Enterprise/Datacenter (On track! ... Most SSDs in development today for both enterprise and Client are TLC)
Cloud/Hyperscale becomes dominate SSD market (On track. They are building tons of SSDs that are not counted in most market statistics)
Formfactors cost optimized for SSD applications
M.2 for all applications (On track. Client is there, Datacenter is work on implementations)
BGA for Phone/ultramobile, Notebooks (On track … Multiple Example Notebooks from multiple SSD vendors in 2017. Then market acceptance in 2018+)

SCM memory has multiple suppliers and technologies for >$1B market (On track … in 2018/19 ... but the winner might not be who you were expecting)

Mark Webb