D. Atienza, EPFL - YINS
C. Ballif, EPFL - Synergy
M. Kayal, EPFL - SmartGrid
J. Lygeros, ETHZ - HeatReserves
C. Moser, EPFL - SHINE
C. Piguet, CSEM - IcySoC
D. Atienza, EPFL - YINS
C. Ballif, EPFL - Synergy
M. Kayal, EPFL - SmartGrid
J. Lygeros, ETHZ - HeatReserves
C. Moser, EPFL - SHINE
C. Piguet, CSEM - IcySoC
D. Atienza, EPFL - YINS
C. Ballif, EPFL - Synergy
M. Kayal, EPFL - SmartGrid
J. Lygeros, ETHZ - HeatReserves
C. Moser, EPFL - SHINE
C. Piguet, CSEM - IcySoC
Christophe Ballif, EPFL Synergy Maher Kayal, EPFL SmartGrid John Lygeros, ETHZ HeatReserves Christophe Moser, EPFL SHINE Christian Piguet, CSEM IcySoC Chair: Mario Paolone, EPFL David Atienza, EPFL
YINS YINS: Energy- and Thermal-Aware Design of Many-Core Heterogeneous Datacenters
Prof. David Atienza ESL EPFL Prof. Luca Benini IIS ETHZ Dr. Edouard Bugnion EC EPFL Prof. Babak Falsafi PARSA EPFL Prof. Lothar Thiele TIK ETHZ Prof. John Thome LTCM EPFL Dr. Fabrice Roudet Eaton Gmbh Marcel Ledergerber Credit Suisse Patrick Segu BrainServe Nano-Tera Annual Meeting, May 20 th 2014 (Lausanne) Information Technologies (IT) Is Ever More Indispensable But Energy-Efficient IT Needed Data Centers Era of knowledge economy and Big Data in science 50% of economic value in developed countries [Economist] Datacenter are key in supply-chain of products/services
Centralized processing with minimal-energy use Computing paradigm to enable energy-proportional services Molecular Dynamics Gene Sequencing Financial Simulations Medical Analytics Weather Prediction Higher Demand + Lower Efficiency: IT Energy Not Sustainable with Current Datacenters! Modern datacenters increase power demands 20 MW Since 2005 - Limited voltage scaling: Trans. smaller, but similar used power Power is beginning to clearly dominate costs in datacenter management
In Switzerland, 3-4% of all electricity, growing at >20% Swiss industry is heavily based on services and requires significant IT support Swiss Energy Strategy 2050: sustainable IT needs to be at its core
B i l l i o n
K i l o w a t t
h o u r / y e a r
2001 2005 2009 2014 2019 0 40 80 120 160 200 240 280 Datacenter Electricity Demands In the US [Energy Star] A Modern Datacenter
17x football stadium, $3 billion 50 million Swiss homes Why Cant Industry Fix This? IT industry/market is horizontal: Per-vendor layer Well-defined interfaces Near-neighbor optimization at best
Big cloud operators (Google, Amazon, Facebook, etc.) Driving the market Do cross-layer optimizations But, Only optimizing their setups Not vendors of technologies And hold data and services hostage
Middleware (data, web services) Application Runtime System (scripting, DSLs) Operating System (resource management) Server (processor, mem, storage, network) Infrastructure (cooling, power) YINS Approach: System-Level Energy-Efficient Integration Balanced and inter-disciplinary consortium to cover datacenter design Multi-scale system engineering: chip to datacenter level (both HW/SW) Synergy of disciplines: electrical eng., mechanical eng. and computer sciences Industrial participation: 3 industrial full partners and 3 third-party partners ESL-EPFL IIS-ETHZ EC-EPFL PARSA-EPFL TIK-ETHZ LTCM-EPFL ~50% less Significant reduction of overall datacenter energy budget with scalable PUE = ~1.0 for all sizes and utilization levels YINS: Project Outcomes and Impact Develop open and green datacenter technologies at system level Adaptive to multiple conditions and industrial needs (e.g., Swiss IT) Critical for the long-term IT independence of Switzerland PUE Utilization 100% 1.03 2.50 1.65 40-80% ~1.00 Homogeneous setup Heterogeneous setups Christophe Ballif, EPFL
Synergy Synergy: Systems for ultra-high performance photovoltaic energy harvesting Christophe Ballif PV-Lab, EPFL CSEM, PV-center Neuchtel What? 12 Systems for ultra-high performance photovoltaic energy harvesting Why? 13 Phase-out of nuclear power production within the next decades
Limited space available. High efficiency approaches
Highly competitive photovoltaics market
Novel approaches
Electronic devices have become more mobile in recent years
Specialty devices
From W to GW The best PV devices for energy generation (1000 W/m 2 ) are also the best devices for energy scavenging (from 10 to 1000 lumen) J = -J ill +J o [e qV/kT -1]+V/R p
J o small (e.g. high bandgap), R p high (no leakage current) High bandgap cells perform better than c-Si in indoor environment (light spectrum and diode law) ! How? 15 Source: Empa Top cell: Narrow-bandgap to harvest visible light Perovskite cells GaAs nanowire cells Wide-bandgap Chalcogenide cells Bottom cell: Wide bandgap to harvest near-infrared light Crystalline/amorphous silicon heterojunction cells Copper indium gallium selenide Tandem solar cells with efficiencies beyond the single-junction limit, i.e. > 30% Consortium 16 Maher Kayal, EPFL
SmartGrid S 3
S 3 Grid: Smart grids, Smart buildings and Smart sensors for Optimized and Secure Management of Electricity Distribution using dedicated microelectronic ICs and real time ICT. Lausanne, 20 th of May 2014 14.10.09 Project Structure and Challenges 19.04.13 19 Smart Grid Smart Buildings Smart Sensors Transmission Network Smart Grid Real time monitoring: Real time power system state estimation and emulation. ICTs dedicated layer. Smart Buildings Control: Demand side management, using intelligent plugs (eSmart) to create a cluster of controllable loads (power distribution). Smart Sensors Local power optimization respecting building occupancy: Zero-power sensors network. Intelligent "human" management of the building by using granular monitoring systems. Active Distribution Network Control 20 Time Reference Monitoring Control State Estimation algorithm Protections Data Concentration Telecom Network PMU Active Distribution Network (ADN) Advanced synchrophasor estimation algorithms for the deployment of PMUs in ADNs.
Resilient PMU synchronization methods for ADNs. Data concentration, time alignment and bad data detection algorithms. Applications for the real-time control and protection of ADNs (e.g., fault location and identification, voltage control, etc.) PMU calibrator PMU Synthesizer KF1 KF2 KF3 Active Distribution Network State Estimation 21 Data Concentration Monitoring Control Time Reference State Estimation algorithm Telecom Network Protections PMU accurate fast (sub-second) reliable Real-Time State Estimation for ADNs Active Distribution Network (ADN) in every network condition (normal and emergency) Recursive SE Kalman Filter ADN state Network topology Measurements Process model Prediction Time Update Estimation Measurement Update 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.9785 0.979 0.9795 0.98 0.9805 0.981 0.9815 Time [s] V o l t a g e
m a g n i t u d e
[ p . u . ]
Measurements True State Kalman Filter True State Measurements Kalman Filter Real Time IP based Communication for Active Distribution Network Dedicated infrastructure. First network deployment to support real time state estimation of several medium voltage buses. Cyber-security architecture. 12.11.12 22 19.04.13 Smart building management 23 Monitoring Acting (ON/OFF) Communication Inertial Loads Specific Loads Battery powered Loads Smart Lights Presence radio sensor Temperature radio sensor Real-Time monitoring Appliances Smart Control John Lygeros, ETHZ
HeatReserves Automatic Control Laboratory, ETH Zrich www.control.ethz.ch HeatReserves: Project Overview John Lygeros Building Energy Management Motivation : Standards: room temperature, air quality, in a given range with predened probability
Cost of energy, green energy Slow building dynamics. Uncertain weather forecasts Objectives : Increase energy eciency Maintain user comfort Computation Weather forecast Occupancy forecast Energy eciency User comfort Building Europe : 40% energy used in buildings Use weather forecast for planning, take into account forecast uncertainty and minimize energy while respecting comfort constraints Buildings for grid ancillary services Traditional electricity grid Inexible, uncertain demand Flexible, deterministic supply Feedback control of supply
Increased in renewables Inexible, uncertain supply
Who can pick up the slack? Flexible, controllable demand? Feedback control of demand through price/other signals Buildings for grid ancillary services Potential of thermal storage for ancillary services Few large or many small buildings Exploit to manipulate electricity demand Sell slack to grid as reserves Facilitate introduction of renewables
Interplay of engineering issues Building climate control for ancillary services Population control of small consumers, Communication requirements
and economic issues Grid value of ancillary services vs. consumer loss of welfare Marketing schemes for consumer participation Other issues: Privacy, Team
Power Systems (ETH-- -PSL) Control Systems (ETH--ACL) Grid Operations (SwissGrid) Building Management (EMPA) Economics (HSG--IWOE) Research questions
Four questions investigated 1. What is the economic value of additional ancillary services to the Swiss grid? 2. How can HVAC systems of oce buildings be controlled to provide ancillary services? 3. How can a large number of household appliances be controlled to provide ancillary services? 4. How can consumers be induced to take part in such demand response schemes? Plus validation on the NEST building (EMPA) Christophe Moser, EPFL
SHINE Optimizing Solar-Hydrogen Production 33 Large electrolyzer Small electrolyzer Optimal electrolyzer Next-generation solar concentrators Solar Driven Water Splitting Unified SHINE Modeling Platform Electromagnetic Wave Propagation (EM) Semiconductor Charge Transport (SC) Heat Transfe r (HT) Reacting Fluid Flow (RFF) Fluid Flow (FF) Electrolyzer Charge Transport (EC) for tomorrow. Christian Piguet, CSEM Andreas P. Burg, EPFL
IcySoC IcySOC RTD 2013 IcySOC RTD 2013 The IcySoC Project A Smart, Ultra-Low-Power Platform for NanoTera Applications IcySOC RTD 2013 IcySOC RTD 2013 The Need for Energy Efficient Electronics
Yesterday Today Tomorrow (autonomous) IcySOC RTD 2013 IcySOC RTD 2013 The Energy Efficiency Challenge 10 12 ops/J 1pJ/op 1GOPS/mW Use high energy when need to work hard Low energy when doing little is good enough Challenge: Energy Proportionality [RuchIBM11] IcySOC RTD 2013 IcySOC RTD 2013 The Mission of IcySoC Provide new solutions and a universal platform for energy-efficient computing for Wearable Health Monitoring Environmental Monitoring The Internet of Things Personal Electronics and Entertainment
IcySOC RTD 2013 IcySOC RTD 2013 The IcySoC Platform Many-core architecture organized in clusters Large number of simple, but efficient processor cores Energy-efficient application-specific accelerators PULP: parallel ultra-low-power platform IcySOC RTD 2013 IcySOC RTD 2013 Why is Low-Power Design Difficult? IcySOC RTD 2013 IcySOC RTD 2013 Approximate Computing: A New Paradigm The Economist, June 2 nd -8 th 2012 Oh, thats near enough: Letting microchips make a few mistakes here and there could make them much faster and more energy-efficient Noisy real world Inputs Perceptual Limitations Approximately Sparse No single Golden Result Multimedia Communications Iterative Self healing Pattern Recognition Application Resilience Web search IcySOC RTD 2013 IcySOC RTD 2013 Smart Energy David Atienza, EPFL YINS Christophe Ballif, EPFL Synergy Maher Kayal, EPFL SmartGrid John Lygeros, ETHZ HeatReserves Christophe Moser, EPFL SHINE Christian Piguet, CSEM IcySoC Chair: Mario Paolone, EPFL