Evaluation result

We demonstrate with above task scheduling and autonomous standby mode transition technology in case of the intruder detection monitor system as a use case of IoT applications (Figure 5.10). The outline of intruder detection evaluation system is shown in Figure 5.11. Here, the evaluation board with multi-sensors are developed and demonstrated. The block diagram and photograph of evaluation board are shown in Figure 5.12 and Figure 5.13 respectively. Here, Renesas MCUs of RL78 (16 bits, max. freq. = 32 MHz, average current = 4 mA) and RX63N (32 bits, max. freq. = 100 MHz, average current = 50 mA) are used.

10m< 5m~10m <5m Motion Sensor (Intruder detect) Sampling Period = Fixed. Infrared Sensor (Approaching object detect)

Sampling Period = 500ms Sampling Period = 50ms Sampling Period = 100ms In conventional case,

- The sampling period of infrared sensor is

no depends on the distance of approaching object. (Predetermined to 50ms by user program)

- The intruder detection is performed at every sampling of motion sensor.

Figure 5.10: Intruder detection system

Power supply & Power measurement board Sensor expansion board Analog F/E +RL78 RX63N 78K0R

Power management unit

Power On/Off Control switch

A

NVRAM (4MB) I/O expansion board Log data output connector wireless communication Expansion board Power evaluation board display Expansion connector Image display Expansion connector

Figure 5.12: Block diagram of evaluation board

RX63N

(MCUhigh)

RL78

(MCUlow)

Radio IF

Sensor

IF

Power

Manager

(FPGA)

USB

I/O IF

NVRAM

68

Task flow of intruder detection system is shown in Figure 5.14. In this use case, the sampling period of the infrared ray (IR) sensor and standby mode of microcontroller are determined with the distance of approaching object by using the autonomous standby mode transition technology, and the judgement of intruder detection is relaxed with the distance of approaching object by using the task scheduling technology, though the sampling period of motion sensor is constant to maintain the detection accuracy as shown in Table 5.2.

Table 5.2: Each state of IR sensor and motion sensor depends on distance of approaching object

>10m 10m~5m 5m> IR sensor sampling period Conv. 50ms fixed. Prop. 500ms 100ms 50ms Motion Sensor sampling period

Conv. Sampling = 30ms fixed. (Judgement= 30ms fixed.)

Prop. 30ms fixed.

(Judgement) (500ms) (100ms) (50ms)

Data sampling at IR sensor (Period=50ms, fixed)

Start

Data sampling at Motion sensor (Period=30ms, fixed )

Intruder detection (Period=30ms, fixed)

Start

Distance calculation from IR sensor data Data sampling at IR sensor

(Period=50/100/500ms) Data sampling at Motion sensor

(Period=30ms, fixed )

Intruder detection (Period=50/100/500ms)

Set IR sampling period and Intruder detection period

In this use case, it is assumed that an intruder has approached once every 1000 s. As the result, around 91 percent power reductions are obtained by adopting the task scheduling and autonomously standby mode transition control technology combination (Proposed-1), where the evaluation period is 1,000 s, and the detection intruder time of 3 s per evaluation period of 1,000 s. (Figure 5.15) In the demonstrated condition, “conventional” corresponds to ASAP scheduling in Figure 5.6 (b), and “proposal” corresponds to Lumped scheduling in Figure 5.6 (c).

The basic idea of normally-off power management in this work is below.

- Lumped scheduling the tasks as much as possible within the deadline period and increasing the standby (idle) period with task scheduling technology.

- Select the optimum standby mode in the extended standby period with autonomous standby mode transition technology.

The energy consumption is more improved with larger number of bundled tasks. Both functions of the task scheduling and autonomous standby mode transition technology work together well.

Conventional

0

100

200

300

400

500

600

Proposed-1 Proposed-2

(Estimated)

Reduction

of 91%

Evauation period = 1,000s

Condition = Detection intruder period of 3s per 1,000s

Proposed-1: Task scheduling + Autonomous standby mode transition technology Proposed-2: Porposed-1 + Nonvolatile memory access control technology

Reduction

of ~1%

E

n

er

g

y

C

on

su

m

p

tion

[J

]

70

The breakdown of energy consumption at this evaluation is shown in Table 5.3 and the selected appropriate standby mode of MCU according to the distance of approaching at this use case is shown in Table 5.4. In conventional case, as IR sensor operates at sampling period of 50 ms, and motion sensor operates at sampling period of 30 ms, these sensors and MCU cannot be powered off. In proposed case, the sampling period of IR sensor and the judgement of intruder detection is relaxed by using the autonomous standby mode transition and task scheduling technologies which are controlled by power manager. The sampling period of IR sensor is

Table 5.3: Breakdown of energy consumption at intruder system evaluation

Conventional Proposed-1 Proposed-2

(Estimated) MCU (RX63N) 161.1 6.9 6.9 MCU (RL78) - 6.9 6.9 IR sensor 306.5 14.7 14.7 Motion sensor 32.7 11.2 11.2 ZigBee 56.7 6.7 6.7 Power Manager - 0.2 0.2 NVRAM - 2.0 1.6 Total 557.0 48.6 48.2

Energy Consumption of each elements [J]

Table 5.4: Selected appropriate standby mode of MCU according to distance of approaching at this use case

>10m 10m~5m 5m> IR sensing

Appropriate standby mode

Conv. IDLE

Prop. Power off Sleep or Power off

IDLE Motion sensing

(Intruder Judgement)

Appropriate standby mode

Conv. IDLE

Prop. Power off Sleep or Power off

extended to 500 ms when the distance of approaching object is over 10 m, and 100 ms when that is 5~10 m. Although the sampling period of motion sensor is fixed to 30 ms to maintain the detection accuracy and these sampling data are stored in NVRAM. However, the judgement of intruder detection is also relaxed to 500 ms when the distance of approaching object is over 10 m, and 100 ms when that is 5~10 m. It means the dead-line period of motion sensor data is 500 ms at > 10 m, and 100 ms at 5~10 m, and the number of bundled tasks is increased and the idle time can be extended. As the results, IR sensor, motion sensor and MCU can be powered off at standby state and energy consumption can be much reduced as shown in Table 5.3. In this evaluation, the power manager is programed with FPGAs.

In addition, we estimated the low-power effects by using newly proposed nonvolatile memory (NVM) access control technology. In case of this demonstration, the sensing data of IR and motion sensors are stored in nonvolatile memory blocks. The information on which data are stored in which NVM block, is stored in the memory use area designation. Thereby, NVM block is turn on only when accessing controlled by NVM access controller. In this demonstration case, the NVM access power reduction of around 18 percent is estimated by using simulation with above function. Here, NVM memory area is divided into 16 blocks and each block is individually power controlled as shown in Figure 5.9. As the result, the total power reduction of around 1 percent is expected in this intruder detection system. In this application, the power consumption is relatively large except for the memory, so the effect is small, but this proposal is a very important technology in the IoT application with low power consumption. For applications that frequent perform memory access such as image recognition, energy improvement by proposed NVM access control technology can be expected.

Here, other use cases are considered. As mentioned above, the energy consumption is more improved with larger number of bundled tasks. In case of Gas sensing, the sampling period is usually set to 20 s. By JIA (The Japan Institute of Architects) inspection regulations, when gas leakage is detected, it is necessary to warn within 60 s. Therefore, the deadline period for task scheduling is set to 60 s, and the number of bundled tasks is 3. So, the energy improvement is estimated to around 20~30 percent which is not so much compared with intruder detection system. Here, in case of intruder detection system, the maximum number of bundled tasks is 16 (=500 ms/30 ms) when the distance of approaching object is over 10 m.

72

5.6. Summary

The low-power multi-sensor system with task scheduling and autonomous standby mode transition control are proposed, and the possibility of power reduction of 91 percent are demonstrated in the use case of the intruder detection monitor system. Furthermore, the low-power effect of around 1 percent by using newly proposed nonvolatile memory access control technology is estimated. These ideas are key candidates applied to IoT applications that require low power consumption.

Chapter 6 Conclusion

This thesis reports emerging embedded nonvolatile memory solution for ultra low power microcontroller systems. This study is presented as measures to address the following three main points:

1. A dual-mode sensing scheme of capacitor-coupled EEPROM cell (Chapter 3) 2. A high-density and high-speed 1T-4MTJ MRAM with voltage offset self-reference

sensing scheme (Chapter 4)

3. A Power Management Scheme with Hierarchical Power Gating and Autonomous Standby Mode Transition Control for Normally-Off Multi-Sensor Network Applications (Chapter 5)

In Chapter 1, the background and objective of this study were prefaced. To realize the embedded nonvolatile memory solution for ultra-low power microcontroller systems, it should be overcome above mentioned challenges, which are low voltage operation, high reliability, high speed access, high density, low standby leakage current. In this thesis, author’s approach to overcome the challenge of embedded the nonvolatile memory and investigate that solution for ultra-low power microcontroller systems.

In Chapter 3, a dual-mode sensing (DMS) scheme of capacitor-coupled EEPROM cell for high speed accessibility and high reliability of write cycle endurance are proposed and estimated. This memory cell combines an EEPROM cell with a DRAM cell, and the cell area penalty is estimated to be less than 10% compared with the conventional EEPROM cell. Using this DMS technique, the signal amplitude on the bit line is increased by 120% at 5 ns after word-line selection, and the sensing speed is enhanced by 36% at the cell current of 15 uA by virtue of the additional charge-mode sensing. Furthermore, the cell current can be decreased by 55% compared with the current-mode only sensing scheme. Therefore, the stress applied to the tunnel oxide of the memory transistor can be relieved by decreasing the programming voltage and shortening the programming time, and it is possible to improve the endurance characteristics. With this memory cell structure and sensing scheme, it is possible to realize high-speed sensing in low-voltage operation and high endurance. The capacitor-coupled EEPROM cell and the DMS scheme are promising candidates for high-performance EEPROM’s.

In Chapter 4, a high-density and high-speed 1T-4MTJ MRAM with voltage offset self-reference sensing scheme is proposed as the method for chip area reduction. A 1Mbit-MRAM with world’s first 1T-4MTJ cell structure has been verified by a 130nm CMOS process technology. By the self-reference sense amplifier with voltage offset scheme, high-speed

74

read access operation of 50MHz@4cycle and tAC=56nsec (Simulation result) has been achieved. In 1T-4MTJ, the active area space in memory array is effectively utilized. The array area reduction of -35.7% become possible with the embedded WWL Driver architecture. By using 1T-1MTJ and 1T-4MTJ cells, on-chip hierarchical memory architecture composed of fast 1T-1MTJ cell for cache memory and small 1T-4MTJ cell for large-capacity memory is feasible. An example of microcontroller design indicates a 20% chip size reduction with keeping a microcontroller performance, by incorporating data memory of 32Mb by 1T-4MTJ cell and fast program memory of 4Mb by 1T-1MTJ cell.

In Chapter 5, the low-power multi-sensor system with power management and nonvolatile memory access control for IoT applications, which achieves almost zero standby power at the no-operation modes, are presented. A power management scheme with activity localization can reduce the number of transitions between power-on and power-off modes with rescheduling and bundling task procedures. In addition, autonomously standby mode transition control selects the optimum standby mode of microcontrollers, reducing total power consumption. We demonstrate with evaluation board as a use case of IoT applications, observing 91 percent power reductions by adopting task scheduling and autonomously standby mode transition control combination. Furthermore, we propose a new nonvolatile memory access control technology, and estimate the possibility for future low-power effect. These technologies are the most promising candidates for future embedded nonvolatile memory solution for ultra low power microcontroller systems.

References

[1] T. Kono, “Embedded Non-Volatile Memory Design for Highly Reliable Applications,” Forum on ISSCC, Feb. 19, 2012.

[2] Y. Terada et al., “A new architecture for the NVRAM – An EEPROM backed-up dynamic RAM,” IEEE J. Solid-State Circuits, vol. 23, no. 1, pp. 86-90, Feb. 1988.

[3] T. Nakayama et al., “A 5-V-only one-transistor 256K EEPROM with page-mode srase,” IEEE J. Solid-State Circuits, vol. 24, no. 4, pp. 911-915, Aug. 1988.

[4] M. Momodomi st al., “An experimental 4-Mbit CMOS EEPROM with NAND-structured cell,” IEEE J. Solid-State Circuits, vol. 24, no. 5, pp. 1238-1243, Oct. 1989.

[5] Y. Terada et al., “120-ns 128K x 8-bit / 64K x 16-bit CMOS EEPROM’s,” IEEE J. Solid-State Circuits, vol. 24, no. 5, pp. 1244-1249, Oct. 1989.

[6] T. Nozaki et al., “A 1-Mb EEPROM with MONOS memory cell for semiconductor disk application,” IEEE J. Solid-State Circuits, vol. 26, no. 4, pp. 497-501, Apr. 1991.

[7] D. Cioaca st al., “A million-cycle CMOS 256K EEPROM,” IEEE J. Solid-State Circuits, vol. SC-22, no. 5, pp. 684-692, Oct. 1987

[8] V. N. Kynett et al., “A 90-ns one-million erase/program cycle 1-Mbit flash memory,” IEEE J. Solid-State Circuits, vol. 24, no. 5, pp. 1244-1249, Oct. 1989

[9] M. McConnell et al., “An experimental 4-Mb flash EEPROM with sector erase,” IEEE J. Solid-State Circuits, vol. 26, no. 4, pp. 484-491, Apr. 1991.

[10] M. Momodomi et al., “A 4-Mb NAND EEPROM with tight programmed Vt distribution,” IEEE J. Solid-State Circuits, vol. 26, no. 4, pp. 492-496, Apr. 1991.

[11] M. Hayashikoshi et al., “A dual-mode sensing scheme of capacitor coupled EEPROM cell for super high endurance,” in Dig. Tech. Papers, 1991 Symp. VLSI Circuits, May 1991, pp. 89-90.

[12] K. Kobayashi et al., “A high-speed parallel sensing architecture for multi-megabit flash E2_{PROM’s,” IEEE J. Solid-State Circuits, vol. 25, no. 1, pp. 79-83, Feb. 1990.}

[13] Y. Terada et al., “High speed page mode sensing scheme for EPROMs and flash EEPROMs using devided bit kine architecture,” in Dig. Tech. Papers, 1990 Symp. VLSI Circuits, June 1990, pp. 97-99.

[14] T. Nakada et al., “Normally-off Computing for IoT Systems,” ISOCC, 2015, pp. 147-148. [15] T. Kono et al., “40nm embedded SG-MONOS flash macros for automotive with 160MHz

76

212-213.

[16] Y. Taito et al., “A 28nm embedded SG-MONOS flash macro for automotive achieving 200MHz read operation and 2.0MB/s write throughput at Tj, of 170°C,” ISSCC, 2015, pp. 132-133.

[17] H. Mitani et al., “A 90nm Embedded 1T-MONOS Flash Macro for Automotive Applications with 0.07mJ/8kB Rewrite Energy and Endurance Over 100M Cycles Under Tj of 175°C,” ISSCC, 2016, pp. 140-141

[18] M. Ueki, et al., “Low-Power Embedded ReRAM Technology for IoT Applications,” in Symp. On VLSI Technology Dig. Tech Papers, pp. T108-T109, 2015.

[19] Y. Hayakawa, et al., “Highly reliable TaOx ReRAM with centralized filament for 28-nm embedded application,” in Symp. On VLSI Technology Dig. Tech Papers, pp. T14-T15, 2015.

[20] Y. Choi, et al., “A 20nm 1.8V 8Gb PRAM with 40MB/s Program Bandwidth,” in ISSCC Dig. Tech. Papers, pp. 46-48, 2012.

[21] C. Villa, et al., “A 45nm 1Gb 1.8V Phase-Change Memory,” in ISSCC Dig. Tech. Papers, pp. 270-271, 2010.

[22] R. Ogiwara, et al., “Highly Reliable Reference Bitline Bias Designs for 64 Mb and 128 Mb Chain FeRAMs,” in Solid-State Circuits, IEEE Journal, pp.1324-1331,2015.

[23] M. Qazi, et al., “A low-voltage 1Mb FeRAM in 0.13μm CMOS featuring time-to-digital sensing for expanded operating margin in scaled CMOS,” in ISSCC Digest of Technical papers, pp. 208-210, 2011.

[24] M. Durlam, et al., “Nonvolatile RAM based on magnetic tunnel junction elements,” in ISSCC Digest of Technical papers, pp. 130-131, 2000.

[25] T. Tsuji, et al., “A 1.2V 1Mbit Embedded MRAM Core with Folded Bit-Line Array Architecture, ” in Symp. VLSI Circuits Dig. of Tech. Papers, pp. 450-453,2004.

[26] N. Sakimura et al, “A 512Kb cross-point cell MRAM,” ISSCC Digest of Technical Papers, pp. 278-279,2003.

[27] Y. Asao et al, “Design and Process Integration for High-Density, High-Speed, and Low-Power 6F2 _{Cross Point MRAM Cell,” in IEDM Technical Digest, pp. 571-574, 2004.}

[28] M. Jefremow et al, “Time-differential Sense Amplifier for sub-80mV Bitline Voltage Embedded STT-MRAM in 40nm,” ISSCC, 2013, pp. 216-217.

[29] T. Andre et al, “ST-MRAM Fundamentals, Challenges, and Applications,” CICC, 2013, pp. 1-8.

[30] C. Kim et al, “A Covalent-Bonded Cross-Coupled Current-Mode Sense Amplifier for STT-MRAM with 1T1MTJ Common Source-Line Structure Array,” ISSCC, 2015, pp.

134-135.

[31] H. Noguchi et al, “A 3.3ns-Access-Time 71.2μW/MHz 1Mb Embedded STT-MRAM Using Physically Eliminated Read-Disturb Scheme and Normally-Off Memory Architecture,” ISSCC, 2015, pp. 136-137.

[32] J. C. Slonczewski, “Current-driven Excitation of Magnetic Multilayers,” J. Magn. Magn. Mater., vol. 159, pp. L1-L7, 1996.

[33] W. H. Butler et al, “Switching Distributions for Perpendicular Spin-Torque Devices Within the Macrospin Approximation,” IEEE Trans. Magn., vol. 48, pp. 4684-4700, 2012.

[34] J. DeBrosse et al., “A Fully-Functional 90nm 8Mb STT MRAM Demonstrator Featuring Trimmed, Reference Cell-Based Sensing,” Custom Integrated Circuits Conference (CICC), pp. 1-3, 2015.

[35] H.Tanizaki et al., “A high-density and high-speed 1T-4MTJ MRAM with Volt-age Offset Self-Reference Sensing Scheme,” in A-SSCC, pp. 303-306, Nov.2006.

[36] T. Y. Kim et al., “A 75MHz MRAM with Pipe-Lined Self-Reference Read Scheme for Mobile/Robotics Memory System,” in A-SSCC, pp. 117-120, Nov.2005.

[37] H. Kano et al., “MRAM with improved magnetic tunnel junction material,” in INTERMAG Conf., p. BB-04, 2002.

[38] R. Scheuerlein et al., “A 10 ns read and write non-volatile memory array using a magnetic tunnel junction and FET switch in each cell,” ISSCC Digest of Technical Papers, pp. 128-129, 2000.

[39] B. N. Engel et al., “A 4-Mb Toggle MRAM Based on a Novel Bit and Switching Method,” in IEEE Transactions Vol. 41, Issue 1, pp. 132-136, 2004.

[40] H. Honigschmid et al., “Signal-Margin-Screening for Multi-Mb MRAM,” in Digest of Technical Papers, pp. 467-476, 2006.

[41] T. Ohsawa et al., “1Mb 4T-2MTJ nonvolatile STT-RAM for embedded memories using 32b fine-grained power gating technique with 1.0ns/200ps wake-up/power-off times,” in VLSI Circuits (VLSIC), pp. 46-47, 2012.

[42] K. Tsuchida et al., “A 64Mb MRAM with Clamped-Reference and Adequate-Reference Schemes,” in Solid-State Circuits Conference Digest of Technical Papers (ISSCC), pp. 258-259, 2010.

[43] C. J. Lin et al., “45nm Low Power CMOS Logic Compatible Embedded STT MRAM Utilizing a Reverse-Connection 1T/1MTJ Cell,” in Electron Devices Meeting (IEDM), pp. 1-4, 2009.

[44] S. Chung et al., “Fully integrated 54nm STT-RAM with the smallest bit cell dimension for high density memory application,” in Electron Devices Meeting (IEDM), pp. 12.7.1 - 12.7.4,

78

2010.

[45] B. H. Krogh, et al., “Cyber-Physical Systems Executive Summary,” CPS Steering Group, IEEE Int. Conf. Cyber-Physical Syst., Mar. 6, 2008.

[46] R. Puri, L. Stok et al., “Keeping Hot Chips Cool,” Proceedings of the 42nd Annu. Design Autom. Conf., 2005, pp. 285–288.

[47] L. Zhao et al., “Gayser-2: The second prototype CPU with fine-grained run-time power gating,” in Proc. 16th Asia and South Pacific Design Autom. Conf., 2011, pp. 87–88.

[48] K. Usami et al., “Design and Control Methodology for Fine Grain Power Gating based on Energy Characterization and Core Profiling of Microprocessors,” in proc. 19th Asia and South Pacific Design Autom. Conf., 2014, pp. 843-848.

[49] L. Benini et al., “A survey of design techniques for system-level dynamic power management,” IEEE Trans. Very Large Scale Integr. Syst., vol. 8, no. 3, pp. 299–316, Jun. 2000.

[50] M. Weiser, et al., “Scheduling for Reduces CPU Energy,” in Proc. 1st USENIX Conf. Operating Syst. Des. Implementation, 1994, Art. no. 2.

[51] F. Yao, et al., “A scheduling model for reduced CPU energy, Foundations of Computer Science,” in Proc. 36th Annu. Symp., 1995, pp. 374–382.

[52] W. Huang, et al., “An optimal speed control scheme sup-ported by media servers for low power multimedia applica-tions,” Multimedia Syst., vol. 15, no. 2, pp. 113–124, 2009. [53] M. E. T. Gerards, et al., “Optimal DPM and DVFS for Frame-based Real-time Systems,”

ACM Trans. Archit. Code Optim., vol. 9, no. 4, pp. 41:1–41:23, 2013.

[54] T. Nakada, et al., “Design Aid of Multi-core Embedded Systems with Energy Model,” IPSJ Trans. Adv. Comput. Syst., vol. 7, no. 3, pp. 37–46, 2014.

[55] T. Nakada, et al., “An Adaptive Energy-Efficient Task Scheduling under Execution Time Variation based on Statistical Analysis,” in Proc. IFIP/IEEE Int. Conf. Very Large Scale Integration Conf., 2016, pp. 1–7.

[56] T. Nakada, et al., “Energy-Efficient Continuous Task Scheduling for Near Real-Time Periodic Tasks,” in Proc. IEEE Int. Conf. Data Sci. Data Intensive Syst., 2015, pp. 675–681. [57] M. Hayashikoshi, et al., “Low-Power Multi-Sensor System with Task Scheduling and

In document Emerging embedded nonvolatile memory solution for ultra low power microcontroller systems (Page 78-102)