TSMC announced on Monday that its customers have started shipping products based on chips made by TSMC using its N7+ (2nd Generation 7 nm with EUV) process technology that uses extreme ultraviolet lithography (EUVL) for up to four layers. The company also said that its clients are on track to tape out chips to be made using N6 node next year.

When compared to N7 (1st Generation 7 nm) that solely relies on deep ultraviolet lithography, TSMC lists its N7+ process as providing a 15% to 20% higher transistor density as well as 10% lower power consumption at the same complexity and frequency. Furthermore, after less than two quarters in production, TSMC is stating that N7+ now matches N7’s yields, which has been used for over a year now.

Use of EUVL enables TSMC to reduce usage of multipatterning technologies when printing highly complex circuits. This also means that TSMC’s EUV tools can offer output power of greater than 250 watts for day-to-day operations while reaching target goals for availability.

Advertised PPA Improvements of New Process Technologies
Data announced by companies during conference calls, press briefings and in press releases
  TSMC
16FF+
vs
20SOC
10FF
vs
16FF+
7FF
vs
16FF+
7FF
vs
10FF
7FF+
vs
7FF
6FF
vs
7FF
5FF
vs
7FF
Power 60% 40% 60% <40% 10% ? 20%
Performance 40% 20% 30% ? same (?) ? 15%
Area Reduction none >50% 70% >37% ~17% 18% 45%

At present, TSMC uses N7+ to produce chips for multiple customers. The most prominent customer of N7+ is Huawei's Hisilicon with the Kirin 990 5G.

TSMC is on track to start risk production of semiconductors using its N6 process technology in the first quarter of 2020 and initiate high-volume production using this node by the end of next year. TSMC’s N6 is a further development of N7 that offers 18% higher transistor density, uses EUVL for up to five layers and enables designers of chips to re-use the same design ecosystem (e.g., tools, IP, etc.), which lowers development costs. By contrast, N7+ uses different design rules, but also provides more benefits than N6 when compared to N7.

Related Reading

Source: TSMC

POST A COMMENT

27 Comments

View All Comments

  • FreckledTrout - Thursday, October 10, 2019 - link

    I was replying to you about the process node and the fact you thought Anton was wrong. Anton was not wrong is what i'm saying. As far as Apple I have no idea but I highly suspect they are on 7nm not the 7nm+ process. Reply
  • melgross - Wednesday, October 09, 2019 - link

    Apparently, Apple chose the N7 Pro second generation process instead. Both have their advantages. Reply
  • Anymoore - Tuesday, October 08, 2019 - link

    I have to wonder if 7nm+ uses this: https://patents.google.com/patent/US9091930B2/en

    They have continued the application at USPTO as well.

    It would increase the die size as was seen with Huawei Kirin 990 5G.
    Reply
  • Oberoth - Wednesday, October 09, 2019 - link

    Is TSMC's 6nm sometimes referred to as 5nm because what was all this news about?
    https://www.techpowerup.com/259536/tsmc-to-begin-m...
    Reply
  • levizx - Thursday, October 10, 2019 - link

    No. 5nm already entered risk production months ago. Reply
  • FreckledTrout - Thursday, October 10, 2019 - link

    The 5nm node is going to be the next real full node change. This is where the magic begins in my opinion. Once we get most features at 5nm we are taking transistor densities that can have CPU and GPU's all in one chip or chiplets on the same package with very little sacrifices. I suspect the dedicated GPU market to retract a ton once we have proper 5nm APU's. Reply
  • Raqia - Thursday, October 10, 2019 - link

    Looking forward to that too as GPUs at the 7nm process are already full fledged and pack very interesting features not found on high power envelope parts:

    https://www.highperformancegraphics.org/wp-content...

    Memory bandwidth will have to go up from the roughly ~30 GB/s on 2x 32bit channels found today on most high end SoCs to support that additional compute at 5nm. There is a rumored move to LPDDR5 shortly and there have been 2x 64bit channel configurations such as on the Snapdragon 805 in the past which will help, and the much higher densities of this process will also allow for much larger caches on die to hide low DRAM bandwidth and latencies better.
    Reply

Log in

Don't have an account? Sign up now