Normally-Off MCU Architecture for Low-power Sensor Node ASP-DAC 2014, Session 1S-3 Jan. 21, 2014 Masanori Hayashikoshi (Speaker), Yohei Sato, Hiroshi Ueki, Hiroyuki Kawai, Toru Shimizu, Renesas Electronics Corporation 1
Outline Introduction Lower power requirements of sensor nodes in smart society Challenges for low-power sensor node Proposal for normally-off architecture for low-power sensor node Summary and Conclusions 2
Outline Introduction Lower power requirements of sensor nodes in smart society Challenges for low-power sensor node Proposal for normally-off architecture for low-power sensor node Summary and Conclusions 3
Paradigm shift for power savings Intermittent system operation by NVRAM will be a promising solution to low-power requirements. SRAM based operation power active standby Large SRAM standby leak time eflash based operation NVRAM based operation <Features> - Fast power-on/off - No DC current during standby power active standby power active eflash start-up standby time Data evacuation to eflash time <Advantages> - Very low power consumption - Reliability improvement of Trs in peripheral circuits 4
Challenge for further power saving in NVRAM system application Normally-off computing technology is enable to cut off the power, except for the component in work truly even in operation as a system, with synergy of NVRAM and power gating technology. Dependence of activity op1 op3 op5 application algorithm op2 op4 op7 op6 op8 scheduling activity Unit A activity Unit B op1 op2 If t<bet, No chance for Power-Off BET t Power On (BET: Break Even Time) op5 op6 op8 op3 op4 op7 time time activity Activity activity Localization If t>bet, Chance for Power-Off Maximize the power-off period with activity localization op1 BET t Power Off op5op6op8 op2 op3 op4 op7 time time 5
Technologies to realize Normally-Off Computing Power-Off as long as possible, as large area as possible, if QOS is satisfied. Power-Off during IDLE-Time of circuits, function blocks, chips, units and systems. Application-independent Computing Platform based on Power-gating, Software-control and OS API. Circuit Technology Device Technology Clock gating DVFS, AVS, etc. Power gating Non-volatile RAM Software Technology Power on-off control, OS, API Normally-Off Computing is a System-level platform technology with combining HW and SW technologies. 6
Outline Introduction Lower power requirements of sensor nodes in smart society Challenges for low-power sensor node Proposal for normally-off architecture for low-power sensor node Summary and Conclusions 7
Cyber-Physical System is a fundamental technology in Smart City The advanced service can be realized with collecting realtime information in the social and natural environments. Data Center Finance, Government, School, Content Vender Music, Visual, Game, Network service, Power Company Remote Meter Reading, Power Supply Control BEMS, HEMS Air-conditioning management, Seculity Gurd, Maintenance Service Vender Gloval Processing <Inter-net> Mobile Internet Smart Grid Power Network Environment monitor Cloud Telecommunication Modem Network Connected Modem Server Power Control Server Home Security Motion, Thermal, Smoke Sensors Door Key Living Information Service Consumer Electronics Control Automated Mater Reading Sensor Camera Air Conditioning Thermal,Humidity Sensor Power Meter Intercom Data Delivery with PC Water Meter Power Monitoring Light Gas leak Detection Gas Meter Regional and local information collection <Sensor-net> Therefore, sensor nodes become used extensively to gather the real-time information. 8
Production volume of Sensor Node Production Volume: ~10B pcs/year This volume will be much increased with the development of cyber-physical systems. It becomes very important how to reduce the power consumption of huge sensor nodes. PC & Server Mobile Phone & Terminal Embedded Sensor-node 0.3B pcs/year 1.5B pcs/year 10B pcs/year (Production Volume: 2011 Data) 9
Outline Introduction Lower power requirements of sensor nodes in smart society Challenges for low-power sensor node Proposal for normally-off architecture for low-power sensor node Summary and Conclusions 10
Conventional Sensor node Sensor-modules are in Normally-On. Microcontroller is in Normally-On or Intermittent. Vcc Input Sensor 0 Input Sensor 1 Input Sensor n Vdd Power Control Data IF CPU (CPU-2) Peripheral POR OSC LDO Memory Sensor-net Interface Sensor network Net work Sensor modules Normally-On Microcontroller Normally-On or Intermittent 11
Challenge for Low-power Sensor nodes Power consumption change of sensor nodes Power Istby ~10uA Power On Overhead 50ms~10s Sensing CPU processing 50us~100ms Standby Breakdown of power consumption of sensor system Wake up (Data communication) Time Challenge for low-power consumption 66% MCU (Stand-by) Telecom 11% 13% Sensor (Always Power-on) MCU (Active) 10% Maximize power-off period. (MCU and Sensor is power on, when necessary) Reduce the active current. 12
Power On/Off Overhead Power On and Off has Power and Time overhead. Normally-off Computing should manage the overhead to reduce total-power. Maximize Power-Off period - Longer Off-time, and More frequently Minimize Power-on energy - Lower active current, and Shorter On-time V(t) I(t) Low-leak (OFF-time) SW-ON Program start Power-ON Low-power (ON-time) Power-ON SW-OFF Power-OFF Normally-off (Longer and More frequently) Power-OFF SW-ON Program start Power-ON Power-ON SW-OFF 13 t t
Outline Introduction Lower power requirements of sensor nodes in smart society Challenges for low-power sensor node Proposal for normally-off architecture for low-power sensor node Summary and Conclusions 14
Normally-Off Architecture for Low-power Sensor node Sensor-modules are in Normally-Off. Microcontroller is in Normally-Off. Low Load High Load Input Sensor 0 Input Sensor 1 Input (a) Vcc-1 Sensor n n Data IF Data IF Vcc-2 n (b) Sensor Control (CPU-1) Sensor modules Normally-Off Normally-On Normally-Off Vcc Power Manager Power Statement Register Vcc-3 Control Sig. (C) Sensor data buffers Vcc-4 Vdd Power Control Data IF Control Sig. (d) CPU (CPU-2) Peripheral POR OSC Memory Sensor-net Interface Microcontroller Normally-Off LDO RTC Sensor network Net work 15
Normally-Off Architecture for Low-power Sensor node (Continue) Challenge is how to control the following key components. Normally-Off Power Manager (a) Power Manager controls power on/off in all units with data of Power Statement Register, and Manages task-level scheduling for activity localization. Sensor Controller (b) Simple processing (e.g. data sampling) which was carried out in the microcontroller, are performed in Sensor Controller. Sensor Data Buffer (c) Sensor data are stored in Sensor Data Buffer. After that, microcontroller performs the processing at once to reduce the number of power-on of microcontroller. 16
Normally-Off Architecture for Low-power Sensor node (Continue) CPU in Microcontroller (d) Proposed architecture is consists of heterogeneous CPUs with Sensor Controller(b) and CPU(d). Sensor required high load task, such as a Image sensor, is directly processed at CPU(d). Task processed at Sensor Controller(b) and CPU(d) respectively should be optimized with task-level scheduling method. Thus, it is possible to reduce the operating frequency of the microcontroller, and to maximize the power-off time of microcontroller. 17
Normally-Off Power Management (1) Autonomous standby mode transition Current microcontroller has some standby modes. Normally-Off Power Manager also supports the Autonomous standby mode transition technology to select optimal standby mode for programmer s usability improvement. Consumption Energy EOH: Energy Over-Head BET: Break Even Time Active Energy Standby Energy Power-Off EOH Stby mode-3 EOH Stby mode-2 EOH Stby mode-1 EOH IDLE EOH BET1 BET2 BET3 BET4 Task cycle 18
Normally-Off Power Management (2) Activity Localization technology Sensing data is buffered in sensor data buffer, and after then, microcontroller is activated and performed the process at once. It is possible to optimize the number of power on/off cycles and decrease power consumption energy. Conventional PG control Hierarchical PG control Sensor MCU Sensor Sensor data buff. MCU Td: deadline Td := Td Tbuf effective deadline Data sampling Data sampling Stand-by Ti: interval Power-on time Tbuf :buffer OH Ti: interval Stand-by Power-on Processing for N inst. bundle time 19
Evaluation Board Evaluation board is consist of, Sensor (Temperature, Brightness, Pyroelectric(Motion)) Microcontroller Other Peripheral circuits Power Supply Brightness sensor LCD Communication connector Pyroelectric (Motion)sensor Temperature sensor Analog connector MCU Memory connector Reset SW LED USB RS-232C E1 debugger 20
Evaluation Results Power consumption energy is reduced by 67%. Energy [mj] 6.00 5.00 4.00 3.00 2.00 Sensor Microcontroller (MCU) Reduction by 67% 1.00 0.00 - MCU: conventional PG control - Sensor: always power-on - MCU: Hierarchical PG control - Sensor: Intermittent PG control [Condition] - 3 sensors (Temp., Brightness, Pyroelectric) - Sampling cycle = 1s - Processing period = 4ms - Room temp. 21
Demonstration of Normally-Off Microcontroller system Demonstrate effectiveness and adaptability of normally-off microcontroller system. Demand transportation system with Normally-Off sensors (Pyroelectric, Camera) are under development as a demonstration with Future University Hakodate. Control Center Location Direction Intelligent Bus Stop System with Normally-Off. Dispatch depending on Requirements Demand transportation system based on Intelligent bus stop 22
Outline Introduction Lower power requirements of sensor nodes in smart society Challenges for low-power sensor node Proposal for normally-off architecture for low-power sensor node Summary and Conclusions 23
Summary and Conclusion Production volume of sensor nodes is much increased with the development of cyber-physical systems. Normally-off architecture of microcontroller for future lowpower sensor node has been proposed to reduce the power consumption of huge sensor nodes. To realize true low-power effects with normally-off computing technology, a co-design of hardware and software technology is very important. In this work, the power consumption energy is reduced by 67%. Normally-Off Computing is a candidate for future low power sensor networks. 24
Thank you for your attention. 25