![Figure 2 Principle of the Linear Variable Optical Filter (LVOF) [2]](/figures/figure-2-principle-of-the-linear-variable-optical-filter-1kbki1pm.webp)
2tJJ
L9U2GL nucpou
A CMOS 128-APS linear array integrated with a LVOF for high-
sensitivity and high-resolution micro-spectrophotometry
Chi Liu*, Arvin Emadi, Huaiwen Wu, Ger De Graaf, Reinoud F. Wolfenbuttel
Faculty of EEMCS, Department for ME/EI, Delft University of Technology,
Mekelweg 4, 2628 CD Delft, The Netherlands
ABSTRACT
A linear array of 128 Active Pixel Sensors has been developed in standard CMOS technology and a Linear Variable
Optical Filter (LVOF) is added using CMOS-compatible post-process, resulting in a single chip highly-integrated high-
resolution microspectrometer. The optical requirements imposed by the LVOF result in photodetectors with small pitch
and large length in the direction normal to the dispersed spectrum (7.2μm×300μm). The specific characteristics of the
readout are the small pitch, low optical signals (typically a photocurrent of 100fA~1pA) and a much longer integration
time as compared to regular video (typically 100μs~63s). These characteristics enable a very different trade-off between
SNR and integration time and IC-compatibility. The system discussed in this paper operates in the visible part of the
spectrum. The prototype is fabricated in the AMIS 0.35μm A/D CMOS technology.
Keywords: Capacitive Transimpedance Amplifier, Correlated Double Sampling, Active Pixel Sensor, IC-compatible
microspectrometers, LVOF, Variable Integration Time
1. INTRODUCTION
1.1 LVOF Microspectrometers
Microspectrometers have found application in many fields due to their small size and their low requirement on the
sample volume. Figure 1 shows a general diagram of a dispersive microspectrometer. A Linear variable Optical Filter
(LVOF) combined with a detector array is a suitable principle for the realization of a high-resolution microspectrometer,
where the LVOF replaces the traditional grating as a dispersion component. A LVOF is actually a resonator cavity with
a wedge shape. It is based on the theory of Fabry-Perot interference and the transmitted wavelength of the LVOF varies
linearly with the cavity thickness. Therefore, a wide spectrum can be covered by the LVOF and no focusing element is
needed. The optical resolution is mainly determined by the reflectance of the highly reflective coatings on the cavity
walls. Complete LVOF fabrication involves CMOS-compatible deposition of the top and the bottom dielectric mirrors
and a tapered layer in between, as shown in Figure 2 [1].
Figure 1 A general diagram for a spectrometer
*c.liu-2@student.tudelft.nl; phone +31 (0)15 2785745; fax +31 (0)15 2785755; ei.ewi.tudelft.nl
Optical Sensing and Detection, edited by Francis Berghmans, Anna Grazia Mignani, Chris A. van Hoof,
Proc. of SPIE Vol. 7726, 772616 · © 2010 SPIE · CCC code: 0277-786X/10/$18 · doi: 10.1117/12.854584
Proc. of SPIE Vol. 7726 772616-1
Downloaded from SPIE Digital Library on 24 Jun 2010 to 131.180.130.109. Terms of Use: http://spiedl.org/terms
Transmittance
1
1
140
1
n
n
nm
a
x
xmm
λ
λ
−
==
−
Figure 2 Principle of the Linear Variable Optical Filter (LVOF) [2]
1.2 Photodetection in LVOF Microspectrometers
A high-quality microspectrometer requires a custom-designed imaging system covered by the LVOF. The combined
limitations of the LVOF design and the imaging system determine the performance of the microspectrometer in terms of
spectral resolving power and light detection limit due to optical throughput (etendue). The spectral resolution of the
microspectrometer is primarily determined by the LVOF design, whereas etendue is limited by the optical design and
imposes the required detection limit of the detector in terms of minimum optical intensity. These parameters determine
the quality of this microspectrometer. Unfortunately, reduced spectrometer dimensions are difficult to combine with high
resolving power and etendue. A reduced optical throughput can be compensated for in a special mode of operation of the
photodetector. Increasing the integration time of the detector is a solution. The spectral resolution should not be
compromised by the photodetector design. The photodetector array specification in terms of element dimensions (pitch)
and number of elements should be sufficient to cover the resolution by the LVOF. The taper angle of the LVOF, as
published in [2], is sufficient for a spectral resolution of 2nm on the wavelength range between 540nm and 720nm over
an LVOF length extending over 1mm. This LVOF will generates a dispersed pattern with a spatial bandwidth of 1/14.4
μm
-1
. According to the Nyquist Sampling Theorem, a pixel pitch is required as small as 7.2μm.
There are commercialized image sensors and linear detector arrays, but their key performances are not optimized for on-
chip photodetection in microspectrometers. There are two problems in the photodetection after the LVOF filtering,
making the commercial image sensors less suitable:
(1) Low luminosity [3]. The function of the filter is to remove most of the input optical power by destructive
interference. Therefore, only a small amount of light reaches the detector array. Thus the imaging system should be
capable of low illumination detection.
(2) CMOS-compatibility. The smallest overall system dimensions are obtained when fabricating a LVOF right on top of
the imaging system with the MOMEMS technology. This is an essential feature in applications such as lab-on-a-chip [4].
Thus a photodetection system with standard CMOS process is preferred for its low cost and high compatibility.
2. SMART CTIA-APS
A CTIA-APS linear array with 128 elements has been designed and fabricated within this framework, where CTIA is
short for Capacitive Transimpedance Amplifier and APS is short for Active Pixel Sensor. This detector array is designed
based the LVOF developed in [2]. The idea is to take advantage of the special optical pattern generated by the LVOF
and to implement an IC-compatible photodetection module suitable for operation at low illumination intensities.
2.1 Detector Array
As discussed, the detector should be qualified in three aspects:
(1) Small pitches along the filter length for high spatial sampling frequency;
(2) Large light-sensitive area, for ensuring maximum sensitivity and SNR;
(3) IC compatibility.
Proc. of SPIE Vol. 7726 772616-2
Downloaded from SPIE Digital Library on 24 Jun 2010 to 131.180.130.109. Terms of Use: http://spiedl.org/terms
AUU
APJb
The first two problems can be solved by applying a linear array of strip pixels. The extension of pixel length should
compensate for the narrow pixel width. The nwell-psubstrate junction is selected for photodetection for an optimized
responsivity in the visible light range. Considering all the specifications, the pixel size was selected at
7.2 300mmmm´ .
2.2 Active Pixel Sensor with CTIA
Image sensors usually apply a junction capacitor as the charge-to-voltage convertor, which is not good for linearity; the
popular 3T-APS has its light sensitive areas in proportional with its junction capacitance, which resulted in a limited
sensitivity. A capacitive transimpedance amplifier is included in each pixel to avoid this problem, as shown in Figure 3
[5] [6].
ΦΦ
_
Φ
T-type Reset Switch
Switch 1 Switch 2
Switch 3
Cint
Figure 3 Capacitive Transimpedance Amplifier (CTIA) with a T-type switch
The introduction of the CTIA brings several benefits:
(1) The link between the light sensitive area and the charge-to-voltage conversion is avoided. The sensitivity can be
improved by enlarging the pixel length and by decreasing the integration capacitor;
(2) The link between the accumulated charge and the integration capacitor is avoided. The linearity can be improved by
implementing the integration capacitor as a poly-to-poly structure;
(3) The amplifier enables the implementation of T-type switches with large off-resistance [7], and thus long integration
time. The readout time can then be traded for higher signal-to-noise ratio;
(4) By means of the virtual ground, the photodiode can be biased with a highly fixed biasing voltage, which reduces the
charge modulation effect dramatically and leads to higher accuracy. The amplifier schematic is shown in Figure 4. A
two-stage circuit is chosen for large output swing. The T-type switch across the input and output of the amplifier results
in an off-resistance much larger than feasible with a single pass switch.
Figure 4 Schematic of the in-pixel amplifier
Proc. of SPIE Vol. 7726 772616-3
Downloaded from SPIE Digital Library on 24 Jun 2010 to 131.180.130.109. Terms of Use: http://spiedl.org/terms
2.3 In-pixel CDS
Correlated Double Sampling as the traditional technique for reducing low-frequency noise is also applied here. The
schematic [5] and the timing chart are shown in Figure 5 .
Figure 5 Schematic and timing chart for the buffered CDS
2.4 Variable Integration Time
Two different controls are applied for the photo detection, fixed integration time control and fixed voltage difference
control. Both control principles are tested. The SNR and the dynamic range for the fixed integration time are [8] [9]:
DR=
max int
10
2
int
20log
()
leakage
r photo leakage
qIt
qI I t
σ
−
++
(1)
SNR=
int
10
2
int
20log
()
photo
r photo leakage
I
t
qI I t
σ
++
(2)
These expressions demonstrate that the SNR increases with integration time while the dynamic range decreases.
Therefore besides the fixed integration time control, the fixed voltage difference control is also introduced to boost both
the dynamic range and the signal-to-noise ratio. The In the fixed voltage difference control, the time consumed by the
CTIA-APS output to increase from V1 to V2 is recorded, with the assistance of a reference clock [10]. This control
principle brings three benefits:
(1) There is always a large amount of photons captured, even for low illumination levels. The photon shot noise will be
the dominant noise source, which means a high SNR detection;
(2) The control principles can be implemented with digital logic circuits and integrated into each pixel simply. Each
pixel can then work independently, needing no other control signals;
(3) The A/D conversion can be finished in the time recording by clock counters;
(4) The dynamic range can be improved by designing the clock and the counter appropriately, unlike in the fixed
integration time control where it is jeopardized by the increasing of integration time. The working principle is shown in
Figure 6.
Proc. of SPIE Vol. 7726 772616-4
Downloaded from SPIE Digital Library on 24 Jun 2010 to 131.180.130.109. Terms of Use: http://spiedl.org/terms
LP" I
LP" III LP" II
LP II
KqO"
mbi.,
mbI!,
eIotI g
eIotI
°[H [Oil '0) [0[(l(
018 [021003 [°[ilU
[8 [021003
[r[[O1U
DOJO [0111110) T[tt!U
a11JJllH
arldlflØuwn,O)
LII 38(13
vilD
UP!POIOqd
>
-
1[1Il8
1nd1O0
01011110] - [[11011] 50]
v1.L3
-
apIpoloqd
8 [nUll I"] U11]SU] VII]
0P1P0101[,1
iaJJllfl and,no UWll03 nIl!] SO] VII]
ap!po[ol[a
Figure 6 Working principle for fixed voltage difference control
2.5 Circuit Diagram
The circuit diagram of the readout is shown in Figure 7 (for simplicity the readout of only four pixels is shown). The
standard timing chart of the pixel operation over one cycle is also presented in Figure 8.
Figure 7 Complete system diagram
Figure 8 Standard timing chart for one pixel operation in one cycle
Proc. of SPIE Vol. 7726 772616-5
Downloaded from SPIE Digital Library on 24 Jun 2010 to 131.180.130.109. Terms of Use: http://spiedl.org/terms