FD-SOI: A process booster for ST-Ericsson’s next generation NovaThor, Part 1

With the recent evolution in smartphone capabilities consumer expectations are rising fast. Ultra-fast multicore Gigahertz processors, stunning 3D graphics, full HD multimedia and high-speed broadband connectivity have become the norm for high-end devices. Consumers expect these features to be delivered in a device that is slim, light and can last for at least as long as their previous phones did. For our customers, the product designers, this translates into requirements for delivering high performance at low power in a cost effective manner. Fully Depleted Silicon On Insulator – or FD-SOI - is a technology that addresses exactly these requirements.

At Mobile World Congress, our CEO Didier Lamouche confirmed during his speech that our next generation NovaThor platform, a successor to the NovaThor L8540, will be using 28nm FD-SOI technology.

FD-SOI is a technology that is available for design today and will allow existing designs in 28nm to benefit today already from significant improvements in performance and power. FD-SOI solves – with less process complexity – scaling, leakage and variability issues to further shrink CMOS technology beyond 28nm.

FD-SOI, like FinFET, is a technology that was initially planned for 20nm nodes and below to overcome traditional bulk CMOS scaling limitations such as high leakage and device variability. However, unlike FinFET, FD-SOI process remains a low-complexity planar process very similar to the traditional CMOS bulk. This allows for a faster process development and ramp-up and an easier design porting for existing designs. The strong collaboration between ST-Ericsson, STMicroelectronics, Leti and Soitec is allows us to already benefit in 28nm from the added value of FD-SOI. The three key benefits realized are leading performance, competitive speed/leakage trade-offs, and optimized power efficiency. This post looks at the performance aspects and a later post will look at the other two benefits.

Leading-edge performance across a wide voltage range

The graph below compares the maximum frequency achievable for a particular critical path of an ARM Cortex™-A9 CPU core implementation, versus the supply voltage Vdd, for a slow corner process (SS) and a worst case temperature.

Each curve represents a specific 28nm process offer.

  • 28HP-LVT is a mobile high performance bulk CMOS process. Targeting high CPU performance mobile applications , these processes are derived from fast process flavors with very thin gate oxide and therefore have a limited Vdd overdrive capability (~1.0V) for reliability reasons
  • 28LP-LVT is a low power bulk CMOS process. Traditionally used for low power mobile applications, LP processes are based on thicker transistor gate oxide supporting a higher voltage overdrive (up to 1.3V).
  • 28FDSOI-LVT is the 28nm FD-SOI process developed by STMicroelectronics. FD-SOI uses a similar gate structure as 28LP, it can also sustain a 1.3V overdrive.

In all process, only low voltage threshold (LVT) transistors are considered. These are the one giving the highest speed performance.

Performance comparison of 28nm technologies

Performance comparison of 28nm technologies

  1. First observation is that FD-SOI at nominal voltages (0.9V for HP, 1.0V for both LP and FD-SOI) gives similar peak performance to HP processes and more than 35% performance improvement compared to LP at same Vdd.
  2. Furthermore, higher Vdd tolerance allows for an extra performance boost in FD-SOI that is not possible with HP processes, resulting in better overall peak performance
  3. At low operating voltages such as Vdd=0.6V, the LP process is either not functional or gives low performance. FD-SOI is equivalent or better than the HP process – but with a much lower leakage and dynamic power consumption as we will see in a later post.
  4. Thanks to lower process variability than any bulk CMOS process, FD-SOI allows even lower operating voltages (down to 0.5V) at frequencies that are useful for non-CPU intensive processes (200MHz-300MHz) e.g. Hardware accelerated audio or video playback.

So, over a large Vdd range (from 0.5V up to 1.3V), FD-SOI comprehensively outperforms existing bulk CMOS processes dedicated to mobile applications. This extra performance gain can be used either to increase peak performance or to operate at a lower Vdd for the same performance, saving dynamic power.

(Link to Part2)

  1. Can you clarify the point about performance at the same VDD?
    At VDD of 1.0V the SOI process produces only 2.0GHz while the 28HP process has 2.4GHz, so it appears the 28HP process has a better frequency at the same VDD.

     
    • Effectively Daniel, at same VDD (and for high VDD values), 28HP process will have a better frequency than FDSOI. However, what really counts in mobile application is not so much the supply level but the overall power consumption (the DMIPS/mW). So even if FDSOI needs an higher VDD supply to reach the same frequency as HP, the HP processes are so leaky compared to the FDSOI process that the overall power consumption remains in favor of FDSOI. Actually, this will be treated in the Part 2 of this blog to be posted soon.

       
  2. When will the L8540 be ready for high volume production ?

     
  3. Does this graph consider substrate bias?

     
  4. Addressed in Part 2, thanks Tammy!

     
  5. Can you explain what a WC temperature is in this case. Is WC referring to speed or power (-40C, 0C, or 85C …)?

     
    • Hello Per,
      I am talking here about worst case temperature for speed.
      Temperature inversion for 28NM bulk and 28NM FDSOI is not occuring for the same voltage. It is around 1.1V – 1.2V for bulk depending of the VT implant, while it is around 700mV- 800mV for FDSOI. It means that above this voltage level, the slowest temperature will be for the highest one (+125C) and below it will be for the lowest one (-40C). Around this threshold voltage , the slowest temperature will be anywhere between -40C and 125C. In the present curves I am really comparing worst case temperature independantly for each processes, so for example the 1.0V points are at 125C for FDSOI and -40C for bulk; while at 0.6V all points are at -40C.
      Hope this answers your question.

       
  6. Hello Valery,
    1. Performance comparison between 28HP and 28SOI at their respective nominal voltages is I guess OK (as opposed to equivalent Vcc), but immediately needs to factor in the quadratic+ jump in power for the differential. (btw, 28HP Vcc is 0.85V not 0.9V).
    28HP has multiple flavors: 28HPL and 28HPM. A more equivalent comparison would be 28HPL with its nominal Vcc of 1.0V. The performance is slightly lower (~10%) and significantly lowered leakage (almost 28LP levels) compared to 28HP, and should compare very favorably in performance to 28SOI, if the curves you show are accurate. 28HPM will also have higher overdrive than 28HP.
    FD SOI at a node when FinFETs are introduced is a better option as parasitic reduction is more pronounced between FD SOI & the FinFET device as opposed to FD SOI vs plain bulk device at 28nm.
    Variability should be an advantage for SOI as you note if the rest of the manufacturing recipe is nailed well which hasnt been the case over the last so many years. The biggest advantage would be the absence of trench isolation related variability seen in bulk CMOS.
    Are you sure about temp inversion at 1.1V-1.2V on 28HP. That looks way too high. You might want to check again.

    In general while FDSOI address some of the manufacturability issues seen in PDSOI it brings in its own challenges.

     
    • Hello Pradeep,
      Actually this “HP” process I used for benchmark is a 28LPH process which effectively exhibits a temperature inversion around 1.1V (it depends slightly of the process corner and the poly bias of the standard cell library considered). Part2 of the blog shows comparison of active power, still showing so advantages with FDSOI. As process maturity is progressing, we will be able to provide soon informations not only based on critical path simulations but on real A9 core silicon measurements.

       
      • So its LPH. The higher Vt device would be the reason then. Painful, but its whats it is. Look forward to the actual A9 silicon measurements. Thanks.

         
  7. This is a really useful blog series, Valery. I’m telling lots of people about it. Do you think you could you include a link to the second part of this blog (http://blog.stericsson.com/blog/2012/05/st-ericsson-general/fd-soi-a-process-booster-for-st-ericssons-next-generation-novathor-%e2%80%93-part-2-2/) at the end of the text? And conversely, put a back-link in the second blog to this one? Thx!

     
  8. Pingback: FD-SOI: A process booster for ST-Ericsson’s next generation NovaThor – Part 2 | mobile technology, mobile architecture | ST-Ericsson Technology Blog

  9. Pingback: Opportunities for next generations smartphones in China | mobile technology, mobile architecture | ST-Ericsson Technology Blog

  10. Pingback: Bringing smart solutions to mass market | mobile technology, mobile architecture | ST-Ericsson Technology Blog

  11. Pingback: Everything gets connected – hello future, here we come | mobile technology, mobile architecture | ST-Ericsson Technology Blog

Leave a Comment