Citation: Byun, S.-J.; Kim, D.-G.;
Park, K.-D.; Choi, Y.-J.; Kumar, P.; Ali,
I.; Kim, D.-G.; Yoo, J.-M.; Huh, H.-K.;
Jung, Y.-J.; et al. A Low-Power
Analog Processor-in-Memory-Based
Convolutional Neural Network for
Biosensor Applications. Sensors 2022,
22, 4555. https://doi.org/s22124555
Academic Editors: M. Jamal Deen,
Subhas Mukhopadhyay,
Yangquan Chen, Simone Morais,
Nunzio Cennamo and Junseop Lee
Received: 24 May 2022
Accepted: 14 June 2022
Published: 16 June 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Communication
A Low-Power Analog Processor-in-Memory-Based
Convolutional Neural Network for Biosensor Applications
Sung-June Byun
1,2
, Dong-Gyun Kim
2,3
, Kyung-Do Park
3
, Yeun-Jin Choi
1
, Pervesh Kumar
1
, Imran Ali
1
,
Dong-Gyu Kim
1
, June-Mo Yoo
1,2
, Hyung-Ki Huh
2
, Yeon-Jae Jung
2
, Seok-Kee Kim
2
, Young-Gun Pu
1,2
and Kang-Yoon Lee
1,2,
*
1
Department of Electrical and Computer Engineering, Sungkyunkwan University, Suwon 16419, Korea;
steven7264@skku.edu (S.-J.B.); exeric@skku.edu (Y.-J.C.); itspervesh@skku.edu (P.K.);
imran.ali@skku.edu (I.A.); rlarlarbrb@skku.edu (D.-G.K.); fiance2@g.skku.edu (J.-M.Y.);
hara1015@skku.edu (Y.-G.P.)
2
SKAIChips, Suwon 16419, Korea; horsnal@skku.edu (D.-G.K.); gray@skaichips.co.kr (H.-K.H.);
yjjung@skaichips.co.kr (Y.-J.J.); skkim@skaichips.co.kr (S.-K.K.)
3
Department of Artificial Intelligence, Sungkyunkwan University, Suwon 16419, Korea; pkd0213@skku.edu
* Correspondence: klee@skku.edu or klee@skaichips.co.kr; Tel.: +82-31-299-4954
Abstract:
This paper presents an on-chip implementation of an analog processor-in-memory (PIM)-
based convolutional neural network (CNN) in a biosensor. The operator was designed with low
power to implement CNN as an on-chip device on the biosensor, which consists of plates of 32
×
32
material. In this paper, 10T SRAM-based analog PIM, which performs multiple and average (MAV)
operations with multiplication and accumulation (MAC), is used as a filter to implement CNN at
low power. PIM proceeds with MAV operations, with feature extraction as a filter, using an analog
method. To prepare the input feature, an input matrix is formed by scanning a 32
×
32 biosensor
based on a digital controller operating at 32 MHz frequency. Memory reuse techniques were applied
to the analog SRAM filter, which is the core of low power implementation, and in order to accurately
grasp the MAC operational efficiency and classification, we modeled and trained numerous input
features based on biosignal data, confirming the classification. When the learned weight data was
input, 19 mW of power was consumed during analog-based MAC operation. The implementation
showed an energy efficiency of 5.38 TOPS/W and was differentiated through the implementation of
8 bits of high resolution in the 180 nm CMOS process.
Keywords:
convolutional neural network; processor-in-memory; AI controller; smart sensing con-
troller; on-chip implementation on a biosensor
1. Introduction
As an artificial intelligence model, CNN is considered to be suitable for classifica-
tion
[1,2]
. It is common to classify the images that are input through photos or image
sensors after dataset-based learning [
3
,
4
]. Since each neuron makes classifications based
on the features of input data, identifying the features between data is the key to CNN
operation. For example, in a model for image recognition, each neuron is divided into
neurons based on red, green, and blue for input data. Neurons with high values for
pixels with red, neurons with high values for pixels with green, and neurons with high
values for pixels with blue have high values. In the case of biosensors, when a disease
is included, neurons related to the disease have high values. In recognition of this CNN
model’s excellent classification ability, it is also applied to the field of disease classification
through the use of a biosensor [
5
,
6
]. Prior to the application of the CNN model, scientists
relied strongly on the data accumulated from the experience of researchers and analysis
of detailed differences between the data. However, CNN-based biosensor systems can
obtain the desired data from unlabeled data through learning. The CNN approach is being
Sensors 2022, 22, 4555. https://doi.org/10.3390/s22124555 https://www.mdpi.com/journal/sensors
