Meeting the Design Challenges of nano
Transcription
Meeting the Design Challenges of nano
1 Meeting the Design Challenges of nano-CMOS Electronics Campbell Millar∗ , Scott Roy, David Cummings, Tim Drysdale, Steve Furber, Doug Edwards, Mark Zwolinski, Andy Tyrrell, Alan Murray, Steven Pickles, Richard O. Sinnott, David Berry and Asen Asenov ∗ c.millar@elec.gla.ac.uk, Device Modelling Group, University of Glasgow, www.nanocmos.ac.uk A BSTRACT CMOS transistor scaling has driven the phenomenal success of the semiconductor industry, delivering faster, cheaper, more functional circuits. 40 nm MOSFETs are in mass production at the 90 nm technology node, and sub-10 nm transistors will be in production by 2018 (in ultra-thin-body SOI or FinFET form)[1]. However, as devices scale, microscopic variations in their atomically granular structure give rise to macroscopically measurable variations between devices[2], [3]. Industry now recognises that such variations represent a major challenge to the scaling and integration of current and future nano-CMOS transistors and circuits, and that it will drive revolutionary changes in the way that future integrated circuits and systems are designed. This Glasgow led e-Science Pilot Project, supporting 11 PDRAs and 7 PhD students, combines the top device, circuit, and system design teams in the UK, with industry players in device manufacture, TCAD, analogue & digital fab, and systems design. We aim to show how e-Science technologies can enable a revolution in the electronics design process[4]. Traditional IC design (Figure 1) uses a hierarchical approach that decouples the device, circuit, and systems in order to manage design complexity. Historically (Figure 2) a single device architecture of fixed size required a single compact model set. However, by the 25nm node, in addition to multiple VT devices co-existing on the same chip, bulk devices will be superseded by fully depleted SOI, ultra thin body SOI, and various forms of multi-gate devices including FinFETs. Atomic scale differences in the structure of devices made on these scales cause measurable, ineradicable Figure 1. Traditional decoupled design hierarchy. Figure 2. nodes. Multiple device architectures for sub 25nm technology Figure 3. Sub-threshold characteristics for an ensemble of 200, 35nm gate length MOSFETs. Typical device is shown inset, with granular potential distribution due to the random distribution of dopants in channel, source and drain. Standard toolflow for some practical design, (90nm, 45nm?) complete, down past place & route, so that we have complete cell, device, interconnect information. Hardware + Software Future Research System ALU Figure 4. Transfer characteristics of an ensemble of 200 SRAM cells constructed from 35nm gate length MOSFETs such as those to the left. Random parameter fluctuations cause device mismatch, with critical impact on circuit yield. variations in their macroscopic parameters. They can no longer be considered nominally identical, and must be treated as a statistical ensemble. Circuits too (Figure 4) show the statistical spread characteristic of such ‘atomicity’. Therefore, multiple compact model sets must be available, both for each device architecture supported on a chip, and statistically within each architecture. A more useful design methodology, pervasively supporting statistical design, also needs to be constructed (Figure 5), but in order to be useful must be cast to a methodology familiar to users. The project uses e-Science technologies to develop such a hybrid design methodology. It will be trialled by academics undertaking research in device, circuit and system simulation, and will help us to understand and Subsystem/ system etc. ALU etc. ALU Bus ALU etc. Bus Synapses ALU Registers etc. Bus Synapses Registers Bus Registers Synapses Synapses Top level timing ; and out-of-foundry yeild, and power yei (either calculated from (ei toggles or by more to detailed analysis) de Manchester (Digtal and Analogue), Glasgow DC, Edinburgh Back Annotated RTL-Synopsys Filters Logic blk. timing timi iming inte Blk. interconnect info. Cell D Cell A Cell F Cell C Cell D Cell B Logic Blocks Cell D Cell F Cell A Cell D Cell B Cell D Cell A Cell F Cell A Cell B Cell D Cell C Cell H Cell F Cell A Cell D Cell C Cell E Cell C Cell A Cell G Cell A Cell F Cell D Cell B Cell D Cell E Cell G Cell E Cell A Cell E Cell B Cell D Cell C Cell A Cell A etc. Bus Synapses Registers Registers Cell G Cell E For each Logic block cre create a ‘card index’ of statistically gen generated blocks Manchester, Southampton, York, Edinburgh, Glasgow Interconnect Cell G Cell H Cell G Modelsim ??? Possibly VHDL-AMS Gate/cell Gate/c e/cell netlists tim Std cell timing Standard Cells A A B B A F B F B B A F A B F A B A A Transistor Transi nsistor netlists A F A B B B For each Standard Cell York, Southampton, Edinburgh create a ‘card index’ of Glasgow (Devices), Glasgow (Circuits) cre sta statistically generated cells cel Spice/Randomspice Aurora SPICE mod models Devices For each transistor family create a ‘card fam index’ of statistically ind generated devices gen Technology info. Techno Figure 5. Responsibility Future Research Fundamental physics phy Glasgow (Devices) Geronimo The nanoCMOS variability aware design flow. forecast the behaviour of next-generation nano-CMOS based systems. Using compact models ensembles which characterise variability in a given technology we will characterise sets of standard cells using grid technology to enable the statistical characterisation of circuits, via Monte-Carlo methods. Cells will be simulated using spice, when fully characterised the behaviour and statistics of the cell library will be modelled using high level design languages (VHDL-AMS). Once Logic blocks/subsystems have been described back annotated RTL will allow designers to asses the impact of device variability on architectural level designs. It is likely that variability will have to be combatted on several levels of the design hierarchy in order to deliver truly robust and fault tolerant designs. R EFERENCES [1] ITRS, “International technology roadmap for semiconductors 2005 edition,” http://www.itrs.net/Links/2005ITRS/Home2005.htm, 2005. [2] G. Roy, A. R. Brown, F. Adamu-Lema, S. Roy, and A. Asenov, “Simulation study of individual and combined sources of intrinsic parameter,” IEEE Transactions on Electron Devices, vol. 53, no. 12, pp. 3063–3070, 2006. [3] A. R. Brown, G. Roy, and A. Asenov, “Poly-si-gate-related variability in decananometer mosfets with conventional architecture,” Electron Devices, IEEE Transactions on, vol. 54, no. 11, pp. 3056–3063, Nov. 2007. [4] nanoCMOS eScience Pilot Project, “http://www.nanocmos.ac.uk.”