publication . Preprint . 2019

Stagioni: Temperature management to enable near-sensor processing for energy-efficient high-fidelity imaging

Kodukula, Venkatesh; Katrawala, Saad; Jones, Britton; Wu, Carole-Jean; LiKamWa, Robert;
Open Access English
  • Published: 30 Dec 2019
Abstract
Vision processing on traditional architectures is inefficient due to energy-expensive off-chip data movement. Many researchers advocate pushing processing close to the sensor to substantially reduce data movement. However, continuous near-sensor processing raises the sensor temperature, impairing the fidelity of imaging/vision tasks. We characterize the thermal implications of using 3D stacked image sensors with near-sensor vision processing units. Our characterization reveals that near-sensor processing reduces system power but degrades image quality. For reasonable image fidelity, the sensor temperature needs to stay below a threshold, situationally determined...
Subjects
free text keywords: Electrical Engineering and Systems Science - Signal Processing, ACM-class: C.1.3
Related Organizations
Download from
64 references, page 1 of 5

380 360 [3] Microsoft Hololens. https://www.microsoft.com/en-us/research/blog/

()trrTaeeeupKm333320246800000 TTjaumncb ()trrTaeeeupKm333240000 TTlhoiwgh oDfifmicley-lit Ionfdfioceor dOauytdligohotr [[54]] PiecEEdnoa.neemrdAaar-sg/zltd-lyoaebe-rlelekEopf&hnfis-idscslDoehiDeam,inrseDn-tterfi.dpoinbRergLu-eo-utpessesaesdilrilne,nfSag-iIrdyn-.nsmrgLiitnveoifcmgioin,rrsowga,Ss-ni2ocetdha0flftrL1--ssD8h.m/..oBrailevortnilneimngnies,Cm/.“aNorrsey.uhrcotutsbpterse:s/a,/”mdIe:EvSbEclEoaglTasrb.anlevni.adoniand.

260 100 200 Ti m30e0 (s)400 500 2800 50 Tim10e0(s) 150 200 [6] fDo.rPleonwa-,cAos.tF, olorewm-bpsokwi,eXrr.oXbuo,tiacnsdaDpp.lMicoaltoionnesy,,”“iBneRncShSm2a0r1ki7ngWoofrkCsNhoNps:

(a) Adaptive to temperature. (b) Adaptive to lighting. New Frontier for Deep Learning in Robotics, 2017.

[7] L. Cavigelli, M. Magno, and L. Benini, “Accelerating real-time embedded Fig. 10: Increasing ambient temperature (left) and/or decreasing scene labeling with convolutional networks,” in Proc. of the ACM 52nd

Annual Design Automation Conference, 2015. ambient illumination (right) pulls TsNteSaPdy away from Tlow and pushes [8] D. IC, “History of 3d stacked image sensors.” http://www.3dic.org/3D

TsCteAaPdy close to Thigh. Stagioni shifts thermal boundaries to smoothly stacked image sensor. adapt to different ambient conditions. [9] R. LiKamWa, Y. Hou, J. Gao, M. Polansky, and L. Zhong, “RedEye:

in ACM SIGARCH Computer Architecture News, 2016. juggling between lighting scenarios. We provide this trace as input [10] Z. Du, R. Fasthuber, T. Chen, P. Ienne, L. Li, T. Luo, X. Feng, Y. Chen,

to our runtime and collect the temperature trace. Fig. 10b shows and O. Temam, “ShiDianNao: Shifting vision processing closer to the

the temperature trace overlaid with Thigh and Tlow. We can observe sensor,” in ACM SIGARCH Computer Architecture News, 2015. the smooth variation of temperature with light intensity. [11] S. Dodge and L. Karam, “Understanding how image quality affects deep

(QoMEX), 2016. 7 CONCLUSION [12] K. Skadron, M. R. Stan, K. Sankaranarayanan, W. Huang, S. Velusamy,

vision, as demonstrated by recent academic and industrial efforts. (TACO), 2004. However, we show that doing so hampers sensor fidelity due [13] A. Kumar, L. Shang, L.-S. Peh, and N. K. Jha, “System-level dynamic

thermal management for high-performance microprocessors,” IEEE Trans to thermal noise, thereby limiting the adoption of near-sensor on Computer-Aided Design of Integrated Circuits and Systems, 2008. processing. Our characterization reveals that immediate drop in [14] J. Donald and M. Martonosi, “Techniques for multicore thermal manage-

observation to design principles for managing sensor temperature Architecture News, 2006.

[15] C. Isci, A. Buyuktosunoglu, C.-Y. Cher, P. Bose, and M. Martonosi, for efficient temperature regulation and high fidelity temperatures, “An analysis of efficient multi-core global power management policies:

64 references, page 1 of 5
Abstract
Vision processing on traditional architectures is inefficient due to energy-expensive off-chip data movement. Many researchers advocate pushing processing close to the sensor to substantially reduce data movement. However, continuous near-sensor processing raises the sensor temperature, impairing the fidelity of imaging/vision tasks. We characterize the thermal implications of using 3D stacked image sensors with near-sensor vision processing units. Our characterization reveals that near-sensor processing reduces system power but degrades image quality. For reasonable image fidelity, the sensor temperature needs to stay below a threshold, situationally determined...
Subjects
free text keywords: Electrical Engineering and Systems Science - Signal Processing, ACM-class: C.1.3
Related Organizations
Download from
64 references, page 1 of 5

380 360 [3] Microsoft Hololens. https://www.microsoft.com/en-us/research/blog/

()trrTaeeeupKm333320246800000 TTjaumncb ()trrTaeeeupKm333240000 TTlhoiwgh oDfifmicley-lit Ionfdfioceor dOauytdligohotr [[54]] PiecEEdnoa.neemrdAaar-sg/zltd-lyoaebe-rlelekEopf&hnfis-idscslDoehiDeam,inrseDn-tterfi.dpoinbRergLu-eo-utpessesaesdilrilne,nfSag-iIrdyn-.nsmrgLiitnveoifcmgioin,rrsowga,Ss-ni2ocetdha0flftrL1--ssD8h.m/..oBrailevortnilneimngnies,Cm/.“aNorrsey.uhrcotutsbpterse:s/a,/”mdIe:EvSbEclEoaglTasrb.anlevni.adoniand.

260 100 200 Ti m30e0 (s)400 500 2800 50 Tim10e0(s) 150 200 [6] fDo.rPleonwa-,cAos.tF, olorewm-bpsokwi,eXrr.oXbuo,tiacnsdaDpp.lMicoaltoionnesy,,”“iBneRncShSm2a0r1ki7ngWoofrkCsNhoNps:

(a) Adaptive to temperature. (b) Adaptive to lighting. New Frontier for Deep Learning in Robotics, 2017.

[7] L. Cavigelli, M. Magno, and L. Benini, “Accelerating real-time embedded Fig. 10: Increasing ambient temperature (left) and/or decreasing scene labeling with convolutional networks,” in Proc. of the ACM 52nd

Annual Design Automation Conference, 2015. ambient illumination (right) pulls TsNteSaPdy away from Tlow and pushes [8] D. IC, “History of 3d stacked image sensors.” http://www.3dic.org/3D

TsCteAaPdy close to Thigh. Stagioni shifts thermal boundaries to smoothly stacked image sensor. adapt to different ambient conditions. [9] R. LiKamWa, Y. Hou, J. Gao, M. Polansky, and L. Zhong, “RedEye:

in ACM SIGARCH Computer Architecture News, 2016. juggling between lighting scenarios. We provide this trace as input [10] Z. Du, R. Fasthuber, T. Chen, P. Ienne, L. Li, T. Luo, X. Feng, Y. Chen,

to our runtime and collect the temperature trace. Fig. 10b shows and O. Temam, “ShiDianNao: Shifting vision processing closer to the

the temperature trace overlaid with Thigh and Tlow. We can observe sensor,” in ACM SIGARCH Computer Architecture News, 2015. the smooth variation of temperature with light intensity. [11] S. Dodge and L. Karam, “Understanding how image quality affects deep

(QoMEX), 2016. 7 CONCLUSION [12] K. Skadron, M. R. Stan, K. Sankaranarayanan, W. Huang, S. Velusamy,

vision, as demonstrated by recent academic and industrial efforts. (TACO), 2004. However, we show that doing so hampers sensor fidelity due [13] A. Kumar, L. Shang, L.-S. Peh, and N. K. Jha, “System-level dynamic

thermal management for high-performance microprocessors,” IEEE Trans to thermal noise, thereby limiting the adoption of near-sensor on Computer-Aided Design of Integrated Circuits and Systems, 2008. processing. Our characterization reveals that immediate drop in [14] J. Donald and M. Martonosi, “Techniques for multicore thermal manage-

observation to design principles for managing sensor temperature Architecture News, 2006.

[15] C. Isci, A. Buyuktosunoglu, C.-Y. Cher, P. Bose, and M. Martonosi, for efficient temperature regulation and high fidelity temperatures, “An analysis of efficient multi-core global power management policies:

64 references, page 1 of 5
Powered by OpenAIRE Open Research Graph
Any information missing or wrong?Report an Issue