Quantum 2D Monte Carlo Simulation of MOSFET Transistors ...

Quantum 2D Monte Carlo Simulationof MOSFET Transistors using

Advanced Architectures

Computer Architecture GroupDepartment of Electronics and Computer Science

Faculty of PhysicsUniversity of Santiago de Compostela (Spain)

Quantum 2D Monte Carlo Simulation of MOSFET Transistors using Advanced Architectures

R. Valın. E-mail: [email protected]

Outline

! The Computer Architecture Group.

! Motivation of the Nanoelectronics Research.

! Parallel 2D Monte Carlo Simulation of NanoelectronicDevices.

! Assessing the Grid for Nanoelectronics.

! Conclusions.

! Future work: An e-Science Infrastructure for Nanoelectronics.



Introduction of the Computer Architecture GroupWhere are we?



Introduction of the Computer Architecture GroupResearch lines

! Signal Processing for Image, Video Coding, andCommunications.

! Software and Hardware for Computer Graphics.

! Compilation Techniques for Parallel Computers.

! Computer Arithmetic.

! Memory Hierarchy Optimization in Irregular Problems.

! Run-time Support for Parallelisation and MemoryImprovement of Irregular Codes.

! Parallel Simulation of Semiconductor Devices.

! Grid and Cloud Computing.



Introduction of the Computer Architecture GroupParallel Simulation of Semiconductor Devices

Simulation Techniques

! Drift-Diffusion.

! Monte Carlo.

! NEGF.

Simulated Devices

! SOI MOSFETs.

! FinFETs.

! Heterostructures.

! Solar Cells.

! Spin-Diode.



Introduction of the Computer Architecture GroupGrid and Cloud Computing

! FORMIGA PROJECT: Integration ofcomputer labs of the university in thees-NGI.

! FORMIGA CLOUD: Development of aCloud infrastructure using the computerlabs of the university.

Objectives:

! Increasing the number of available resources for oursimulations.

! Learning about new technologies and analysing possibleapplications.



Motivation of the Nanoelectronics ResearchMOSFET Transistor

! MOSFET is one of the basiccomponents of the IntegratedCircuits now and during the last 50years.

! Moore’s Law: 2x transistors everytwo years.

! In this period we had an amazingevolution of the electronics.

! Faster processors, mobiles, newelectronic devices.



Motivation of the Nanoelectronics ResearchMOSFET Transistor



Motivation of the Nanoelectronics ResearchEnd of Traditional Scaling Era

Challenges

! High Speed.

! Integration.

! Power Save.

The International Technology Roadmap for Semiconductors

(ITRS): Guideline that establishes the challenges on the design ofnew devices. The 22 nm technological node is the end of the BulkTechnology (ITRS 2010).



Motivation of the Nanoelectronics ResearchEnd of Traditional Scaling Era

How can Moore’s Lawcontinue?

! Strained Silicon.

! Hi-K.

! Multigate Devices.

! SOI Devices.



2D Monte Carlo Simulation of Nanoelectronic DevicesTechnology for Computer-Aided Design

For technical and economicalreasons Technology ComputerAided Design (TCAD) toolscomplete experimentaldevelopment techniques.



2D Monte Carlo Simulation of Nanoelectronic DevicesMonte Carlo Method

! Numerical method that solves the BTE. Semi-classicaltransport theory.

∂f /∂t + v ·∇rf + k ·∇kf = ∂f /∂t|coll

! The solution of BTE is obtained from the simulation of thecharge carrier transport in the nanoelectronic device.

! Stocastic selection of the flight times and scatteringmechanisms.

! SPMC and EMC.



2D Monte Carlo Simulation of Nanoelectronic DevicesOpenMP Parallelisation of 2D MC Simulator

Why do we need parallelism?

! MC simulators are highdemanding computationalapplications.

! Studying the physical behaviourof a device for 1 ps requiresaround 1 or 2 hours ofcomputational time.

! Typical simulations are 30 pslength.

! Variability studies requireshundreds or thousands ofsimulations.

Why OpenMP?

! 92% of the systems of theTOP500 list have multicoreprocessors (November 2010).73% 4 cores and 19% 6 cores ormore.

! De facto standard forshared-memory programming.

! Quite easy programminglanguage.



2D Monte Carlo Simulation of Nanoelectronic DevicesParallelisation of a 2D Multisubband MC Simulator

! The 2D Multisubband MCSimulator enables us to simulatethe quantum confinement effecton the perpendicular directionto the transport plane.

! BTE is solved in the transportdirection.

! Schrodinger equation is solvedin the confinement direction.




! Execution time has beenreduced by a factor 7 with 8cores.

! We can obtain simulationresults faster.

! This allows us to reduce thedevelopment time of newmodels.




Studies of parameter variations require a lot of simulations.




0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

200

400

600

800

1000

1200

1400

VD (V)

I D (A/m

)

tSiO2 = 0.35 nm

tHfO2= 4.47 nm

UniformtSiO

2 = 1.05 nm

tHfO2= 3.77 nm

1050 1100 1150 1200 1250 1300 13500

5

10

15

20

ID (A/m)

Freq

uenc

y

IontSiO

2 = 0.35 nm

tHfO2= 4.47 nm

tSiO2 = 1.05 nm

tHfO2= 3.77 nm

mean

uniform

! Studies of parameter variations require a lot of simulations.

! More resources are necessary. Could be the Grid useful?



Assessing the GridIntroduction

! Starting point: FORMIGAPROJECT (2007).

! The creation of the EGI based onNGIs is giving a boost to the NGIdevelopment.

! The grid infrastructure of eachcountry will be run by NationalGrid Initiatives.

! The Spanish NGI (es-NGI) issupported by the Spanish Networkfor e-Science.



Assessing the GridAccessing to the resources

! eng.vo.ibergrid.eu VO belongs toIBERGRID infrastructure.

! It was created in 2010 to supportthe jobs of the following areas:architecture and, electrical,electronics and automaticengineering.

Initial tasks:

! Assessing the advantages of thisinfrastructure for nanoelectronics.

! Simplifying the submission andmotorization of the jobs.



Assessing the GridServices

! Information Service: provides thestate of the resources to theresource broker.

! Resource Broker: submits jobs tothe resource centres.

! VOMS: stores the informationabout VOs belonging to es-NGI.

! Storage: Distributed between theresource centres and available forall VO.

! File Catalogue: Localises thestored files.

Moreover, thisinfrastructure relies onmonitoring and accountingservices.



Assessing the GridScheme of the VO infrastructure

! Based on GLite middleware.

Information obtained fromlcg-infosites command(2010).

# Cores Res. Centre

1378 PIC

1616 IFCA

848 IFIC

284 CIEMAT

340 CESGA

148 UNICAN

22 UNIZAR

Total #Cores 4636



Assessing the GridSMNanoS Description

Command line user interface may be a problem for MOSFET VOusers. A Python application has been developed to submit andmonitor jobs.

Functionality levels.



Assessing the GridTest Simulation

! We have simulated a job collection with 5 jobs for testing theresource centres.

! Each job simulates a 2 ps length stationary state of a 2DDGSOI MOSFET.

! The execution script of each job saves the execution time,cpuinfo and kernel characteristics of the WN in the SE.

! These simulations were submitted to different resource centresthat support our VO.

! Simulation results enable us to evaluate the influence of theresource centres heterogeneity on the execution time.



Assessing the GridSimulation Results

Execution time of each job of each resource centre.




Execution time of a job collection = the slowest time of the jobsbelonging to the collection. Execution time deviation of the jobcollections could be over 50%.




# Res. Centre CPU Architecture

PIC Xeon L5420 2.50 GHz x86 64

Xeon L5530 2.40 GHz x86 64

IFCA Xeon E5345 2.33 GHz x86 64

IFIC Xeon E5420 2.50 GHz x86 64

CIEMAT Opteron270 2.0 GHz i686

UNICAN PentiumD 3.0 GHz x86 64

CESGA Pentium4 3.20 GHz i686

Processor models of the WNs for each resource centre.



Conclusions

! Scaling of nanoelectronic devices requires new materials andtechnologies.

! It is necessary to simulate new devices considering advancedphysical models.

! Usually, scaled devices require more complex models andtherefore, more computational time.

! Parallel simulators enable us to develop new models and getsimulation results faster.

! The strong development of multicore processors hasencouraged us to adopt OpenMP as the used parallelprogramming language.

! For example, the parallelisation with OpenMP of a 2DMSB-MC allows us to reduce by 7 the total execution timeusing 8 cores.



Conclusions

! Thanks to the parallel simulators we can get the results ofparameter variations faster. However, more available resourcesare required.

! Grid infrastructures could be a good option to increase thenumber of available resources. But several drawbacks have tobe overcome:

• Heterogeneous resources produce until a 50% difference on theexecution time of the simulations.

• Easier technologies for the final users are required.Authentication, submission and monitoring services use archaiccommand line systems.



Future work: An e-Science Infrastructure for

NanoelectronicsTidying up the nanoelectronic simulations

Nanoelectronics requires:

! Access to differentinfrastructures.

! Run different simulatorsmade with differentprogramming languages.

! Data managementsystems.

! Visualisation tools.




NanoelectronicsProposed Infrastructure

Useful for other research fields. Bioinformatics, Chemistry, etc.




Nanoelectronics



Quantum 2D Monte Carlo Simulationof MOSFET Transistors using

Advanced Architectures

Thank you for your attention!



Quantum 2D Monte Carlo Simulation of MOSFET Transistors ...

Documents

Transcript of Quantum 2D Monte Carlo Simulation of MOSFET Transistors ...