Process technology is extremely complex today, and getting more and more complex by the minute. The billions of dollars invested in each process node essentially insures that it will have to be used for years to come to get back that investment. It not only needs to get back that investment, but provide more funds to start R&D on the next series of nodes that will come down the line. It has only been a couple of years since the introduction of multiple 14nm processes from Intel and Samsung, as well as the 16nm node from TMSC. We are already moving towards an introduction of 10nm parts from these manufacturers in bulk starting next year. So have these manufacturers gotten their money worth out of their current processes?
Kinam Kim, President of Samsung Electronics’ Semiconductor Business, discloses the latest process advances from his division.
Part of that answer somes in the form of Samsung's latest product. Samsung is announcing the availability of a new 11nm FinFET process that looks to be a pretty extensive optimization of the company's 14nm FF. The new process promises 15% better performance and 10% chip area reduction at the same power consumption as the older 14nm FF. The idea here is to further improve upon their 14nm process all the while retaining the economics of it. This process exists separately from the latest 10nm LPP which can be considered a full jump from the previous 14nm. 11nm LPP will be primarily aimed at midrange and high end products, but will not reach the full scaling and performance of the 10nm LPP product.
This "little steps" philosophy has been around for ages, as AMD utilized it for most of their existence when they owned their own Fabs. Other companies have done the same by including small improvements over the lifetime of the process so that the final product is signficantly better in terms of yield, transistor switching speed, and thermal dissipation. Samsung looks to be doing this with their 11nm process by providing all those little steps of improvement from 14nm.
The second part of the announcement is that Samsung has announced their 7nm process using EUV. Samsung had previously announced their 8nm process, but it still relies upon multi-patterning immersion litho. Samsung has been testing their 250 watt EUV source with fairly good results. The company is quoted as to processing over 200,000 wafers since 2014 and has achieved an 80% yeild on 256 Mb SRAM. This is somewhat impressive, but still not ready for primetime. SRAM features highly consistent structures and is typically one of the first complex chips tested on a new process.
Samsung is offering orders now of its 11nm line and it will be very interesting to see who jumps on board. I would not expect AMD to transfer their designs to 11nm, as a tremendous amount of reworking and validating are required. Instead we will see AMD going for the 10nm node with their Zen 2 based products while continuing to produce Ryzen, Vega, and Polaris at 14nm. Those that will be taking advantage of 11nm will probably be groups pushing out smaller products, especially for the midrange and high end cell phone SOCs.
10nm LPP is expected in early 2018, 8nm LPP in 2019, and finally Samsung hopes for 7nm to be available in 2020.
this will be interesting to
this will be interesting to see how this shakes out if glofo can get its homegrown 7nm process up or they screw up again and have to license Samsung process tech again
11nm is in fact an optimized
11nm is in fact an optimized 14nm process and something tells me they chose this number because TSMC is calling 12nm to their optimized 16nm process. These classifications are getting more and more silly :facepalm:
GF should license Samsung’s
GF should license Samsung’s 11nm process and continue to work on their 7nm process and AMD is going to be using TSMC also so GF apparently is operating as close to full capacity anyways. I’m sure now that AMD can make its yearly wafer agreement with GF and still require excess capacity for its GPUs with TSMC. AMD’s Zeppelin dies production alone will probably be enough by itself to surpass AMD’s yearly wafer production agreement with GF allowing AMD to hedge their bets with other fabs should AMD need to and that 14nm process that GF licenses from Samsumg is probably very similar to Samsung’s new 11nm process and maybe AMD should look into updating it’s Ryzen production to 11nm while it trys to get it’s Zen 2 up and running on GF’s in house 7nm node.
AMD’s first generation Zen/Epyc server/workstation products are still going to remain in production for a few more years so AMD can meet its extended product availability commitments to the server/workstation/Ryzen Pro market based on the first generation Zen micro-arch and that’s still where the majority of AMD’s server revenues are going to come from. And even after Zen2 becomes available there will still be years of first generation Zen product availability as the server/workstation/pro markets expect that extended support and any server market movement onto Zen2 will be dependent on the workstation/server OEMs’ longer process of certifying any new CPU micro-arch anyways.
The way AMD has priced its Epyc platform products even AMD’s consumer CPU SKUs are not as affordable on a price per core basis if you compare Threadrippers 16 core 1950X SKU($999) against AMD’s 24 core 7401P SKU($1075) that only costs $76 dollars more than the Threadripper SKU.
AMD’s Server/Workstation SKUs are going to be so popular that even GF is going find it hard to meet its Zeppelin die/wafer demands from AMD on CPU production alone. So that Samsung 11nm process tweak of its 14nm node that GF licenses will offer AMD a chance to get more dies per wafer an a little bit higher clocks across all of its product lines that currently use thet Zeppelin modular die. So maybe a little tick tock update cadence before the Zen 2 products come online under a brand new untested GF 7nm node late next year into 2019.
Maybe AMD should have GF License Samsung’s 11nm for any new first generation Zen/Zeppelin die stepping and try that out for Ryzen production for a little higher clocks to better compete with Intel on the clock rates if that 11nm Samsung process can net enough gains so Ryzen can reach clock parity with Intel’s K series offerings.
Josh, slight error in your
Josh, slight error in your article:
AMD is on record that they are going to skip 10nm and Zen2 will be on GloFo’s 7nm.
For AMD’s sake, I hope GF can
For AMD's sake, I hope GF can deliver.