IEEE Solid-State Circuits Magazine - Fall 2017 - 53

the image input, which, along with
compression, reduces the sensor
bandwidth by 10× [67]. Such a sen-
sor can also reduce the computation
and energy consumption of the sub-
sequent processing engine [48], [68].

Summary
Machine learning is an important
area of research with many prom-
ising applications and opportuni-
ties for innovation at various levels
of hardware design. The challenge
is to balance the accuracy, energy,
throughput, and cost requirements.
Since data movement dominates
energy consumption, the primary
focus of recent research has been
to reduce the data movement while
maintaining accuracy, throughput,
and cost. This means selecting archi-
tectures with favorable memory hier-
archies like a spatial architecture and
developing data flows that increase
data reuse at the low-cost levels of
the memory hierarchy. Joint design
of algorithms and hardware can be
used to reduce bit-width precision,
increase sparsity, and compression
to further reduce the data movement
requirements. Mixed-signal circuit
design and advanced technologies
can be used to move the computa-
tion closer to the source by embed-
ding computation near or within the
sensor and in the memories.
Finally, designers should also con-
sider the interactions between the
different levels of design. For instan-
ce, reducing the bit width through
hardware-friendly algorithm design
enables reduced precision process-
ing with mixed-signal circuits and
nonvolatile memories. Reducing the
cost of memory access with advanced
technologies could also result in more
energy-efficient data flows.

Acknowledgments
Portions of this article contains ex-
cerpts from the invited paper "Hard-
ware for Machine Learning: Challenges
and Opportunities" that appeared at
the 2017 IEEE Custom Integrated Cir-
cuits Conference [69]. The work was
done in collaboration with Yu-Hsin

Machine learning is an important area of
research with many promising applications and
opportunities for innovation at various levels
of hardware design.
Chen, Joel Emer, Amr Suleiman, Tien-
Ju Yang, and Zhengdong Zhang.

References

[1] "Complete visual networking index (VNI)
forecast," Cisco, June 2016.
[2] R. Szeliski, Computer Vision: Algorithms
and Applications. Springer Science & Busi-
ness Media, 2010.
[3] M. Price, J. Glass, and A. P. Chandrakasan,
"A 6 mW, 5,000-word real-time speech
recognizer using WFST models," IEEE
J. Solid-State Circuits, vol. 50, no. 1, pp.
102-112, 2015.
[4] R. Yazdani, A. Segura, J.-M. Arnau, and
A. Gonzalez, "An ultra low-power hard-
ware accelerator for automatic speech
recognition," in Proc. MICRO, 2016, pp.
1-12.
[5] T.-C. Chen, T.-H. Lee, Y.-H. Chen, T.-C.
Ma, T.-D. Chuang, C.-J. Chou, C.-H. Yang,
T.-H. Lin, and L.-G. Chen, "1.4μW/chan-
nel 16-channel EEG/ECoG processor for
smart brain sensor SoC," in Proc. Symp.
VLSI, 2010, pp. 21-22.
[6] K. H. Lee and N. Verma, "A low-power
processor with configurable embedded
machine-learning accelerators for high-
order and adaptive analysis of medical-
sensor signals," IEEE J. Solid-State Circuits,
vol. 48, no. 7, pp. 1625-1637, 2013.
[7] N. Dalal and B. Triggs, "Histograms of ori-
ented gradients for human detection," in
Proc. Conf. Computer Vision and Pattern
Recognition, 2005, pp. 886-893.
[8] D. G. Lowe, "Object recognition from lo-
cal scale-invariant features," in Proc. Int.
Conf. Computer Vision, 1999, pp. 1150-
1157.
[9] N. Cristianini and J. Shawe-Taylor, An Introduction to Support Vector Machines
and Other Kernel-Based Learning Methods. Cambridge, U.K.: Cambridge Univ.
Press, 2000.
[10] R. E. Schapire and Y. Freund, Boosting:
Foundations and Algorithms. Cambridge,
MA: MIT Press, 2012.
[11] Y. LeCun, Y. Bengio, and G. Hinton, "Deep
learning," Nature, vol. 521, no. 7553, pp.
436-444, May 2015.
[12] A. Suleiman, Y.-H. Chen, J. Emer, and
V. Sze, "Towards closing the energy
gap between HOG and CNN features
for embedded vision," in Proc. IEEE Int.
Symp. Circuits and Systems, 2017, pp.
444-447.
[13] Y. LeCun, K. Kavukcuoglu, and C. Farabet,
"Convolutional networks and applica-
tions in vision," in Proc. IEEE Int. Symp.
Circuits and Systems, 2010.
[14] S. Ioffe and C. Szegedy, "Batch normaliza-
tion: Accelerating deep network training
by reducing internal covariate shift," in
Proc. Int. Conf. Machine Learning, 2015,
pp. 448-456.
[15] V. Nair and G. E. Hinton, "Rectified linear
units improve restricted boltzmann ma-
chines," in Proc. Int. Conf. Machine Learning, 2010, pp. 807-814.
[16] Y. LeCun, L. Bottou, Y. Bengio, and P. Haff-
ner, "Gradient-based learning applied to

document recognition," Proc. IEEE, vol.
86, no. 11, pp. 2278-2324, Nov. 1998.
[17] V. Sze, Y.-H. Chen, T.-J. Yang, and J. Emer,
"Efficient processing of deep neural net-
works: a tutorial and survey," arXiv Pre-
print, arXiv:1703.09039, 2017.
[18] A. Krizhevsky, I. Sutskever, and G. E. Hin-
ton, "ImageNet classification with deep
convolutional neural networks," in Proc.
Neural Information Processing Systems
Conf., 2012, pp. 1097-1105.
[19] K. Simonyan and A. Zisserman, "Very
deep convolutional networks for large-
scale image recognition," in Proc. Int.
Conf. Learning Representations, 2015.
[20] C. Szegedy, et al. "Going deeper with
convolutions," in Proc. Conf. Computer
Vision and Pattern Recognition, 2015,
pp 1-9.
[21] K. He, X. Zhang, S. Ren, and J. Sun, "Deep
residual learning for image recognition,"
in Proc. Conf. Computer Vision and Pattern
Recognition, 2016, pp. 770-778.
[22] O. Russakovsky, J. Deng, H. Su, J.
Krause, S. Satheesh, S. Ma, Z. Huang,
A. Karpathy, A. Khosla, M. Bernstein, A.
C. Berg, and L. Fei-Fei, "ImageNet large
scale visual recognition challenge," Int.
J. Computer Vision, vol. 115, no. 3, pp.
211-252, 2015.
[23] C. J. B. Yann LeCun and C. Cortes. The
MNIST database of handwritten digits.
[Online]. Available: http://yann.lecun
.com/exdb/mnist/
[24] M. Horowitz, "Computing's energy prob-
lem (and what we can do about it)," in
Proc. IEEE Int. Solid-State Circuits Conf.,
2014, pp. 10-14.
[25] M. Mathieu, M. Henaff, and Y. LeCun,
"Fast training of convolutional networks
through FFTs," in Proc.  Int. Conf. Learning
Representations, 2014,
[26] C. Dubout and F. Fleuret, "Exact accelera-
tion of linear object detectors," in Proc.
European Conf. Computer Vision, 2012,
pp. 301-311.
[27] A. Lavin and S. Gray, "Fast algorithms for
convolutional neural networks," in Proc.
Conf. Computer Vision and Pattern Recognition, 2016, pp. 4013-4021.
[28] M. Sankaradas, V. Jakkula, S. Cadambi,
S. Chakradhar, I. Durdanovic, E. Cosat-
to, and H. P. Graf, "A massively paral-
lel coprocessor for convolutional neu-
ral networks," in Proc. ASAP, 2009, pp.
53-60.
[29] V. Sriram, D. Cox, K. H. Tsoi, and W. Luk,
"Towards an embedded biologically-in-
spired machine vision processor," in Proc.
Field-Programmable Technology, 2010, pp.
273-278.
[30] S. Chakradhar, M. Sankaradas, V. Jak-
kula, and S. Cadambi, "A dynamically
configurable coprocessor for convolu-
tional neural networks," in Proc. IEEE Int.
Symp. Computer Architecture, 2010. pp.
247-257.
[31] V. Gokhale, J. Jin, A. Dundar, B. Martini,
and E. Culurciello, "A 240 G-ops/s mobile
coprocessor for deep neural networks,"
in Proc. Conf. Computer Vision and Pattern Recognition Workshop, 2014. pp.
682-687.

IEEE SOLID-STATE CIRCUITS MAGAZINE

FA L L 2 0 17

53
http://www.yann.lecun
Table of Contents for the Digital Edition of IEEE Solid-State Circuits Magazine - Fall 2017
