The VLSI Symposia – one on technology and one on circuits – are among the most influential in the semiconductor industry. Three hugely important papers were presented – one on 14nm FD-SOI and two on 10nm SOI FinFETs – at the most recent symposia in Honolulu (9-13 June 2014). In fact, three out of four papers in the Highlights Sessions covered SOI devices for the 10 and 14nm nodes.
There were so many great SOI-based papers that we’re going to cover this conference in two posts. This post covers the three big 14nm FD-SOI and 10nm FinFET papers. Summaries and abstracts of all the others will be covered in Part 2 (click here to read Part 2). Please note that as of this posting date, the papers are not yet available from the IEEE Xplore site – but they should be shortly.
Top SOI Highlights from the Symposium on VLSI Technology
2.3: 14-nm FDSOI Platform Technology for High-Speed and Energy-Efficient Applications. O. Weber et al. (STMicroelectronics, CEA-LETI, IBM)
This is the big paper we’ve been waiting for – the one that indicates 14nm FD-SOI should match the performance of 14nm bulk FinFETs. We still don’t have a side-by-side FD-SOI v. bulk FinFET comparison, as there is scant data at comparable leakage on bulk FinFETs at 14nm publicly available with which to compare. But based on what they’ve been seeing and some extrapolation, the FD-SOI technology developers see the figures presented in this paper as a big win. We’ve already seen hints of this in a recent ASN piece (click here to see that one) showing 14nm FD-SOI matching 14nm bulk for performance and coming in at a much lower cost. Now in terms of performance, here’s the VLSI paper detailing the FD-SOI side of the story.
The authors confirm a scaling path for FD-SOI technology down to 14nm, using strain-engineered FD-SOI transistors. Compared to 28-nm FDSOI, this work provides an 0.55x area reduction from scaling and delivers a 30% speed boost at the same power, or a 55% power reduction at the same speed, due to an increase in drive current and low gate-to-drain capacitance. Using forward back-bias, an additional 40% dynamic power reduction for ring oscillators is experimentally demonstrated. Moreover, a full single-port SRAM is described, including a 0.081 μm2 high-density bitcell and two 0.090 μm2 bitcell designs used to address high-performance and low-leakage/low Vmin requirements.
TEM of an FD-SOI nMOS transistor, showing gate-to-drain capacitance components and experimental values. From 14-nm FDSOI Platform Technology for High-Speed and Energy-Efficient Applications (O. Weber et al., STMicroelectronics, CEA-LETI & IBM)
2.2: A 10nm Platform Technology for Low Power and High Performance Application Featuring FINFET Devices with Multi Workfunction Gate Stack on Bulk and SOI. K.-I. Seo et al. (Samsung, IBM, ST, GF, UMC)
This paper covers the first-ever demonstration of FinFET technology suitable for 10-nm CMOS manufacturing. Targeting low-power and high-performance, it offers the tightest contacted poly pitch (64 nm) and metallization pitch (48 nm) ever reported on both bulk and SOI substrates. A 0.053 μm2 SRAM bit-cell – and this part was on SOI – was reported with a low corresponding static noise margin of 140 mV at 0.75 V. The team developed intensive multi-patterning technology and various self-aligned processes with 193i lithography to overcome optical patterning limits. A multi-workfunction gate stack provides Vt tunability without the variability degradation channel dopants induce.
Projected scaling trend, featuring the tightest contacted poly pitch (CPP=64 nm) and metallization pitch (Mx=48 nm) ever reported, on both bulk and SOI substrates. From A 10nm Platform Technology for Low Power and High Performance Application Featuring FINFET Devices with Multi Workfunction Gate Stack on Bulk and SOI, by K.-I. Seo et al. (Samsung, IBM, ST GF, UMC)
2.4: Strained Si1-xGex-on-Insulator PMOS FinFETs with Excellent Sub-Threshold Leakage, Extremely-High Short-Channel Performance and Source Injection Velocity for 10nm Node and Beyond, P. Hashemi et al. (IBM, GlobalFoundries, MIT)
The authors demonstrated high performance (HP) s-SiGe pMOS pMOSsfinFETs with Ion/Ieff of ~1.05/0.52mA/μm and ~1.3/0.71mA/μm at Ioff=100nA/μm at VDD=0.8 and 1V, extremely high intrinsic performance and source injection velocity. Compared to earlier work, an optimized process flow and a novel interface passivation scheme, result in ~30% mobility enhancement and dramatic sub-threshold-swing reduction to 65mV/dec. They also demonstrate the most aggressively scaled s-SiGe finFET reported to date, with WFIN~8nm and L G~15nm, while maintaining high current drive and low leakage. With their very low GIDL-limited ID, min and more manufacturing-friendly process compared to high-Ge content SiGe devices, as well as impressive Ion~0.42mA/μm at Ioff =100nA/μm and gm, int as high as 2.4mS/μm at VDD=0.5V, s-SiGe FinFETs are strong candidates for future HP and low-power applications.
TEM images of the most aggressively scaled SiGe FinFET reported to date with a fin width of ~8nm and gate length of ~15nm. From Strained Si1-xGex-on-Insulator PMOS FinFETs with Excellent Sub-Threshold Leakage, Extremely-High Short-Channel Performance and Source Injection Velocity for 10nm Node and Beyond, P. Hashemi et al. (IBM, GlobalFoundries, MIT)
There were also two rump sessions held during the conference, which were co-chaired by Soitec CTO Carlos Mazure. The SOI ecosystem was well-represented, the rooms were packed and the debate lively.
Rump Session 1: Who gives up on scaling first: device and process technology engineers, circuit designers, or company executives? Which scaling ends first – memory, or logic? Panelists: M. Bohr, Intel; M. Cao, TSMC; J. Chen, Nvidia; S-H Lee, Hynix; T-J King Liu, UC Berkeley; K. Nii, Renesas: R. Shrivastava, Sandisk; H. Jaouen, STMicroelectronics; E. Terzioglu, Qualcomm
The take-away here is that the majority of panelists and attendees see company executives giving up on scaling in the face of rising costs.
Rump Session 2: 450 mm, EUV, III-V, 3D; All in 7nm? Are you serious?! Panelists: W. Arnold, ASML; R. Gottscho, Lam Research; K. Hasserjian, AMAT; S. Iyer, IBM; C. Maleville, Soitec; A. Steegen, IMEC
The general consensus was that 3D integration is needed and will be adopted at the 7nm node due to delays and the high cost of the EUV and III-V, and the lack of 450mm wafer supply and support.