ARTICLE

https://doi.org/10.1038/s41467-019-10282-1 OPEN

Microring resonator-assisted Fourier transform

spectrometer with enhanced resolution and large
bandwidth in single chip solution
S.N. Zheng1,2, J. Zou1,3, H. Cai2, J.F. Song4, L.K. Chin1, P.Y. Liu1, Z.P. Lin 1, D.L. Kwong2 & A.Q. Liu1
1234567890():,;

Single chip integrated spectrometers are critical to bring chemical and biological sensing,
spectroscopy, and spectral imaging into robust, compact and cost-effective devices. Existing
on-chip spectrometer approaches fail to realize both high resolution and broad band. Here we
demonstrate a microring resonator-assisted Fourier-transform (RAFT) spectrometer, which is
realized using a tunable Mach-Zehnder interferometer (MZI) cascaded with a tunable
microring resonator (MRR) to enhance the resolution, integrated with a photodetector onto a
single chip. The MRR boosts the resolution to 0.47 nm, far beyond the Rayleigh criterion of
the tunable MZI-based Fourier-transform spectrometer. A single channel achieves large
bandwidth of ~ 90 nm with low power consumption (35 mW for MRR and 1.8 W for MZI) at
the expense of degraded signal-to-noise ratio due to time-multiplexing. Integrating a RAFT
element array is envisaged to dramatically extend the bandwidth for spectral analytical
applications such as chemical and biological sensing, spectroscopy, image spectrometry, etc.

1 School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798, Singapore. 2 Institute of Microelectronics, A*STAR

(Agency for Science, Technology and Research), Singapore 138634, Singapore. 3 College of Science, Zhejiang University of Technology, Hangzhou 310023,
China. 4 College of Electronic Science and Engineering, Jilin University, Changchun 130012, China. Correspondence and requests for materials should be
addressed to H.C. (email: caih@ime.a-star.edu.sg) or to L.K.C. (email: chin0062@e.ntu.edu.sg) or to A.Q.L. (email: eaqliu@ntu.edu.sg)

NATURE COMMUNICATIONS | (2019)10:2349 | https://doi.org/10.1038/s41467-019-10282-1 | www.nature.com/naturecommunications 1

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10282-1

O
ptical spectrometer is a critical instrument for spectrum demonstrate 0.47 nm resolution in ~90 nm spectral range with
analysis in various applications such as chemical and low power consumption (35 mW for MRR and 1.8 W for MZI).
biological analysis, environment monitoring, remote The single-chip integrated RAFT spectrometer shows simple
sensing in satellites, hyperspectral imaging, etc. Conventional design and easy package capability to enable compact and robust
spectrometers are based on free space optical engineering tech- spectrometers for various spectral analytical applications.
nology, which are usually bulky and expensive benchtop instru-
ments. Silicon fabrication and integration technology is a
Results
powerful platform to realize chip-scale spectrometer with high
Resolution enhancement with a microring resonator. Figure 1a
compactness, high compatibility, and low cost1–6. Most on-chip
shows schematic of the proposed RAFT spectrometer consisting
spectrometers are based on dispersive elements such as arrayed
of a thermally tunable MRR and MZI operating at the funda-
waveguide grating (AWG)7–10 and planar concave grating7,10–12,
mental quasi-transverse electric (quasi-TE) mode. A broadband
which are quite similar to the conventional grating-based coun-
light is firstly butt-coupled into the input waveguide of the MRR.
terparts. Others exploit the characteristics of photonic devices to
Only the wavelengths satisfying the resonance condition of the
disperse light such as photonic crystal13,14 and random photonic
MRR will be transmitted to the drop port of the MRR leaving a
structures15, and there are also other approaches such as sta-
series of dips in its throughput port30, as illustrated in Fig. 1b.
tionary ring resonators array16 and speckle pattern reconstruction
Here, for simplicity, we choose the smallest resonance wavelength
by spiral waveguides17. Although these on-chip spectrometers can
of the MRR in the detected wavelength range to make the fol-
achieve a relatively high resolution, their main drawback is
lowing discussion and assume λ0 as the initial resonance wave-
scalability for large bandwidth due to large number of detection
length. When heater 1 is activated by an external voltage, the
channels. For instance, AWG based on-chip spectrometer8 has
resonance position of the MRR will shift to λr, inducing a relative
demonstrated a resolution of 0.2 nm (δλ), but it needs 50 (N)
wavelength shift as Δλ = λr − λ0. Figure 1c shows schematic of
detection channels to obtain a bandwidth of 10 nm (λBW = δλ·N).
the resonance wavelength shift of the MRR when different
Such a huge number of detection channels not only causes the
heating power is applied to heater 1. Here, three tuning states are
complexity of the device with N photodetectors (PD), but also
displayed. The shift Δλn is proportional to the heating power
greatly degrades the signal-to-noise ratio (SNR).
applied to heater 1. Then the following tunable MZI will retrieve
Fourier-transform (FT) spectrometers can overcome such
each filtered spectrum from the MRR at each tuning state by
limitations in dispersive optical spectrometers to achieve high
analyzing the output time-domain interferogram detected by the
resolution and high SNR. FT spectrometer on silicon platform
integrated PD. By combining all the retrieved spectra, the original
has been demonstrated using microelectromechanical
input spectrum is obtained (Fig. 1d). The tunable MZI is designed
technology18,19 with comparable performance to the conventional
to be symmetric with length of 2.46 cm for each arm to achieve a
and bulky counterparts. However, it still requires moving parts
resolution R of ~20 nm with moderate power consumption. Free
and cannot be integrated with on-chip light sources and PD,
spectral range (FSR) of the MRR is designed to be larger than the
which reduces its robustness. Besides, the resolution is relatively
resolution value R of the tunable MZI. Resonance wavelength of
low because of the limited traveling range of the actuator. Other
the MRR can be tuned by a value as small as the linewidth, i.e.,
on-chip FT spectrometers include stationary-wave integrated FT
the FWHM which is 0.15819 nm in our experiment. Thus, the
spectrometers (SWIFTS)20,21 and spatial heterodyne spectro-
resolution can be improved from R to FWHM. Therefore, the
meters (SHS)22–27, which are based on spatial interferograms. In
final resolution δλ of the FT spectrometer assisted by an MRR can
SWIFTS, only a single stationary Mach–Zehnder interferometer
be dramatically enhanced.
(MZI) is used. A resolution of 4 nm with a bandwidth of 96 nm is
demonstrated21. On the other hand, SHS utilizes stationary MZI
array to retrieve the input spectrum from a set of under-sampled Thermal tuning. Both MRR and MZI are thermally tunable by
discrete spatial interferogram. Although a relatively high resolu- exploiting TO effect. The simulation results on heat transfer in
tion can be achieved (~0.045 nm), the need of many MZIs (e.g., silicon-on-insulator (SOI) waveguide with TiN heater can be
32) increases the device size and complexity22. Moreover, large found in Supplementary Note 1. Isolation trenches are exploited
number of detection channels are required for both methods, to improve heating efficiency (Fig. 2a–c). With heating power P in
resulting in a low SNR. FT spectrometers can also adopt thermo- heater above, the static temperature of the waveguide T can be
optic (TO) effect to obtain temporal interferogram28,29 with a written as T0 + kTP according to the Green’s functions31, where
demonstrated resolution in several nanometers. However, high T0 is the initial temperature and kT is heating efficiency
resolution is hindered by the limited optical path length and depending on device materials, structures and dimensions. The
refractive index modification in a silicon chip. As a result, it static temperature T is linearly proportional to the heating power
remains challenging to develop an optical spectrometer that is P and the heating efficiency with isolation trenches is 1.6 times
integrated with PD onto a single chip, achieving high resolution, that without isolation trenches according to simulation results
large bandwidth, and high SNR. (Supplementary Fig. 2a). Therefore, with isolation trenches
In this paper, we demonstrate a microring resonator-assisted adoption, the heating efficiency can be effectively improved. The
FT (RAFT) spectrometer, which is realized using a thermally experimental results of influences of thermal isolation trenches on
tunable photonic MZI, cascaded with a tunable microring reso- heating efficiency are presented in Supplementary Note 3. The
nator (MRR) to enhance the resolution and integrated with a PD heating efficiency can be improved to maximal 12 times
onto a single chip. The final resolution depends on the tuning according to experimental results.
resolution of the resonance wavelength of the MRR, which is in In the following analysis, due to the waveguide of MRR and
subnanometer level due to the ultra-narrow linewidth of the MZI working at the fundamental mode, we will only consider the
resonance peak. Hence, the resolution is dramatically boosted by TO tuning induced effect on effective index neff of the
the MRR far beyond the classic Rayleigh criterion of the FT fundamental quasi-TE mode. In practice, some effects should
spectrometer without resorting to large optical path difference be considered. Firstly, the effective index neff of silicon (Si)
(OPD) of the MZI. Compared to existing FT approaches, the waveguide has a strong wavelength dispersion. Secondly, the TO
RAFT spectrometer requires only a single channel to achieve high coefficient (TOC) has a nonlinear behavior during thermal
resolution and large bandwidth, allowing high SNR. We tuning. The thermal expansion due to temperature excursion also

2 NATURE COMMUNICATIONS | (2019)10:2349 | https://doi.org/10.1038/s41467-019-10282-1 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10282-1 ARTICLE

b
a Throughput

ds ut
d pa ughp
Bon Thro PD
er 2
Heat FWHM
er 1

Intensity
Heat
put
Out
MZI Drop
FSR

MRR Drop

Input
Wavelenght

c MRR output spectra d Retrieved spectrum

 = 0  = 1  = n

Transmission
Transmission

Transmission

Intensity
   

Fig. 1 Microring resonator-assisted Fourier-transform spectrometer. a Schematic of the RAFT spectrometer consisting of an MRR and an MZI both with
heaters on top integrated with a PD. b Schematic transmission spectra of the MRR. Free spectral range (FSR) is the separation between adjacent dips
(peaks) and full width half maximum (FWHM) denotes the linewidth of the dips (peaks). c Schematic of output filtered spectra from the drop port of
the MRR at three tuning states (0 < Δλ1 < …… < Δλn < FSR), which are denoted by lines with different styles (colors). d Schematic retrieved spectra at
different tuning states (denoted by lines with different styles (colors)) by the tunable MZI. The original input spectrum (denoted by black dotted line) can
be retrieved by combining all the retrieved spectra

induces waveguide length change ΔL. Besides, the fabrication factor of MRR. For a given MRR, T(v) is a constant for a certain
variance will induce effective index difference (δn) and imbalance wavelength. H(v0) and G(v0) are wavelength-dependent correc-
(δL) between two arms29. The parameter values of waveguide tion factors for imperfect beam splitters and optical losses,
dispersion, TO effect and thermal expansion are presented in respectively. The output power consists of a constant portion B
Supplementary Table 1. Hence, the effective index change Δneff (v0)Ii(v0) and a modulated portion B(v0)Ii(v0)cos(2πv0δ/c). The
should be modified to modulated portion constitutes the interferogram where the
Δneff ¼ Δneff ðv; ΔTÞ  δnðvÞ; ð1Þ intensity changes with OPD. Thus, for a broadband input source,
taking only the modulated portion, the output power intensity is
where ΔT = T − T0 with T0 = 300 K. The total arm length expressed as
difference is expressed as
ΔL ¼ ΔLðΔTÞ  δL: ð2Þ Z
þ1

Io ðτÞ ¼ BðvÞIi ðvÞ cosð2πvτÞdv; ð4Þ

The expressions of Δneff (v, ΔT), δn(v), and ΔL(ΔT) are
presented in Supplementary Note 2. According to the above 1

analysis, the effective index change is proportional to temperature

excursion ΔT (Fig. 2d). where τ = δ/c. When FT is performed to Eq. (4), the input
The resonance wavelength of the MRR is expressed as intensity can be retrieved as
λ
λr ¼ λr0 þ r0  Δneff ; ð3Þ Z
þ1
ng 2
Ii ðvÞ ¼ Io ðτÞ cosð2πvτÞdτ: ð5Þ
where λr0 is the initial resonance wavelength and ng is the group BðvÞ
0
index. The resonance wavelength is proportional to Δneff, thus
proportional to temperature excursion ΔT. Taking account of waveguide dispersion, temperature depen-
For the tunable MZI, the OPD between two arms varies with dent TOC and thermal expansion in Eq. (4), we obtain
the heating power P applied to heater 2 residing above the upper
arm as shown in Fig. 1a, which will result in a different output Z
þ1
intensity for a different P. With a monochromatic source Ii(v0) 1
(v0 = c/λ0) as the input of RAFT spectrometer whereby c is the Io ðΓÞ ¼ BðuÞ½Ii ðuÞ cosðφðuÞÞ cosð2πuΓÞdu; ð6Þ
1 þ ξ1
speed of electromagnetic wave in vacuum, the output power Io(δ) 1
can be expressed as B(v0)Ii(v0)(1 + cos(2πv0δ/c))32, where δ is
OPD. The coefficient B(σ0) can be expressed as 0.5H(v0)G(v0)T where the definitions and values of the parameters u, ξ1, and г are
(v0). The factor T(v) is the wavelength-dependent transmission presented in the Supplementary Note 2. Thus, the modified input

NATURE COMMUNICATIONS | (2019)10:2349 | https://doi.org/10.1038/s41467-019-10282-1 | www.nature.com/naturecommunications 3

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10282-1

Without trench With trench

d 2.380
a
Trench
Heater Heater 2.375

2.370

Effective index
MRR
2.365
MRR
Cladding Cladding
2.360
165 °C
b Heater
160 2.355
MRR
SiO2 120
2.350
280 300 320 340 360 380 400
80
Temperature (K)
40
e 700
120 OPD
24
Si substrate Resolution 600
100
197 °C 500

Resolution (cm–1)
c Heater
80

OPD (µm)
MRR 180 400
SiO2 60
140 300

100 40 200

60 20 100
Isolation trench
24 0 0
Si substrate
0 400 800 1200 1600
Power (mW)

Fig. 2 Silicon-on-insulator (SOI) waveguide thermal tuning. a Schematic top views of a tunable MRR without and with isolation trenches. Simulated static
temperature distribution in the cross-section of a tunable MRR with constant heating power P on the heater above b without isolation trenches and c with
isolation trenches. d Relation between effective index of the quasi-TE mode and temperature. e Relation between OPD, resolution of the tunable MZI, and
heating power on MZI heater 2

spectrum in Eq. (5) can be expressed as eases the requirement on maximal OPD for the tunable MZI. The
Z
þ1 bandwidth can be further improved by paralleling the RAFT
2ð1 þ ξ 1 Þ element array with each designed specifically for a certain
Ii ðuÞ ¼ Io ðΓÞ cosð2πuΓÞdΓ: ð7Þ spectral range.
BðuÞ
0

Finally, the original input spectrum is reconstructed by

transforming u to v, MRR characterization. Figure 3a shows the false-colored optical
uv
micrography of the fabricated RAFT spectrometer. Figure 3b
v¼1þξ 0 þv0
Ii ðuÞ ! Ii ðvÞ:
1 ð8Þ shows the SEM image of the MRR without SiO2 upper cladding,
while Fig. 3c shows its final image with isolation trenches and TiN
B(σ) can be obtained through experimental power calibration. heater. Isolation trenches near MRR are exploited to improve
The resolution of the tunable MZI R is given by 1/Δ where Δ is heating efficiency and reduce thermal crosstalk between the MRR
the maximum OPD. OPD equals to Δneff (L + ΔL). Therefore, heater and MZI heater. Figure 3d shows the optical micrography
OPD is proportional to the heating power applied to heater 2, of a waveguide-coupled Ge-on-SOI PD.
while the resolution value decreases with increasing heating The experiment setup for MRR characterization is illustrated in
power (Fig. 2e). Simulations at different conditions (Supplemen- Supplementary Fig. 17a. Based on the transmission spectrum
tary Table 2) show the resolution of the tunable MZI can be from throughput port of the MRR (Fig. 4a), the resonance
improved either by increasing the arm length and/or increasing wavelengths are λon1 = 1528.256 nm, λon2 = 1555.776 nm, and
the heating efficiency (Supplementary Fig. 10). λon3 = 1584.296 nm. The measured FSR is approximately 28 nm
The frequency information of the input spectrum of the and the linewidth (FWHM) at 1528.256 nm is ~0.15819 nm with
tunable MZI can be extracted by performing fast FT (FFT) to the a quality factor (Q) of approximately 9661. The tuning power
output interferogram (intensity changes with the applied electric consumption is 1.23 mW nm−1 with a maximum estimated
power on MZI heater 2). Since the MRR prefilters the input temperature change of 188.8 K and thus the MRR heater
spectrum to sparsely spaced wavelength components, the tunable efficiency is around 5.5 × 103 K W−1. Subsequently, the transmis-
MZI can differentiate the wavelength components if its resolution sion spectra within one FSR are monitored as shown in Fig. 4b
value is smaller than the FSR. The MRR resonance wavelength while the applied voltage on heater 1 is increased from 0 to 4.4 V.
can be shifted by applying electric power on the MRR heater 1 The experimental data of resonance wavelength and the heating
with a tuning value as small as the FWHM. Thus, the final power on heater 1 can be well fitted with a linear equation as
resolution of the RAFT spectrometer δλ is dramatically enhanced shown in Fig. 4c. Hence, we assume a linear relation between λr
compared to the designed resolution of the tunable MZI R, which and heating power P on heater 1.

4 NATURE COMMUNICATIONS | (2019)10:2349 | https://doi.org/10.1038/s41467-019-10282-1 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10282-1 ARTICLE

a
Throughput

MRR Output
MZI
Input
Photodetector

b c d
Isolation trench

In
Ge
Heater

1 µm 2 µm 5 µm

Fig. 3 Fabricated RAFT spectrometer. a False-colored optical micrography of a RAFT spectrometer after Ge epitaxy growth for PD. b SEM image of an MRR
without SiO2 upper cladding. c SEM image of a tunable MRR with isolation trenches and TiN heater. d Optical micrography of a waveguide-coupled Ge-on-
SOI photodetector

a –40 b –37 c 1618

–45 –42 1614

Resonance wavelength (nm)

–50 1610
–47
on1
Power (dBm)

Power (dBm)

on2 1606
–55 on3 –52
1602
–60 –57
0V 1598
–65 –62 1.0 V
2.0 V 1594
–70 –67 2.8 V
1590
4.4 V
–75 –72 1586
1526 1541 1556 1571 1586 1601 1616 1582 1588 1594 1600 1606 1612 1618 0 5 10 15 20 25 30 35
Wavelength (nm) Wavelength (nm) Electric power (mW)

Fig. 4 MRR characterization. a Transmission spectrum from the throughput port of MRR. The three resonance wavelengths are denoted as λon1, λon2, and
λon3, respectively. b Transmission spectra within one FSR at different applied voltages on heater 1. c Relation between resonance wavelength and heating
power on heater 1. The error bars denote S.D.

Single wavelength characterization. The experiment setup for are improved with increasing heating efficiency kT shown in
RAFT spectrometer characterization is shown in Supplementary Supplementary Fig. 10a, and/or with increasing arm length L
Fig. 17b. A tunable laser source (TLS-1: Santec TSL-510) is used shown in Supplementary Fig. 10b.
for single wavelength characterization. Firstly, heating power is We test the resolution of the RAFT spectrometer using TLS-1.
applied to heater 1 above the MRR to induce a resonance Here, we employ three resonance peaks (λon1 < λon2 < λon3) of the
wavelength shift of Δλ = 3.38 nm compared to the initial static MRR to filter the input source. For simplicity, we define a
state, thus the MRR has one resonance wavelength at 1584.620 detuning wavelength dλ as λoff–λon indicating the difference
nm. Then, TLS-1 with wavelength set at 1584.620 nm and power between off-resonance wavelength λoff and on-resonance wave-
of 6 mW is fed into the RAFT spectrometer. Figure 5a shows the length λon. We compare the retrieved power intensity of λon and
detected interferogram from MZI output port. The intensity λoff after FFT. The power of TLS-1 is set at 8 mW. The MRR is
changes with the applied power on heater 2. More than 80 per- tuned to Δλ = 3.38 nm. Figure 5b shows the retrieved spectra
iods are observed with a maximum OPD of approximately when λon3 = 1584.620 nm and the value of dλ is set to be 0.3 and
128.354 µm, which corresponds to a theoretical resolution of 0.47 nm, respectively. One can see that when dλ = 0.3 nm, the
77.91 cm−1 (19.32 nm at 1584.62 nm). The power consumption retrieved power ratio between the on-resonant wavelength λon
of MZI is approximately 11.185 mW π−1 with heater heating and the detuned off-resonant wavelength λoff equals to 6.85 dB,
efficiency kT = 13.8 K W−1. Hence, the estimated maximum while when dλ = 0.47 nm, the retrieved power ratio is increased
temperature excursion is ΔT = 24.9 K. The calibration of absolute to 10.10 dB. Similarly, the retrieved power ratios at dλ = 0.47 nm
optical frequency v, γ2, and ξ1 and calculations of kT and ΔT are for the other two resonance wavelengths (i.e., λon1 = 1528.488 nm
presented in Supplementary Note 2. The simulation results of and λon2 = 1556.020 nm) are 14.77 and 16.46 dB, respectively. To
relation between OPD, the resolution of the tunable MZI-based effectively filter out the detuned off-resonant components into the
FT spectrometer and heating power on heater 2 on different drop port, we define the retrieved power ratio should be larger
conditions are presented in Supplementary Note 2. OPD is pro- than 10 dB, i.e., the minimum MRR tuning value is 0.47 nm.
portional to the heating power while the resolution value Hence, the resolution of the RAFT spectrometer is defined as
decreases with increasing heating power. The OPD and resolution 0.47 nm. Figure 5c shows the retrieved spectra with TLS-1 set at

NATURE COMMUNICATIONS | (2019)10:2349 | https://doi.org/10.1038/s41467-019-10282-1 | www.nature.com/naturecommunications 5

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10282-1

a 4 b 9 d = 0.3 nm 9 d = 0.47 nm
c 9.0
1528.488 nm
1556.02 nm
1584.62 nm
7.5 1532.488 nm
1560.02 nm
3 1588.62 nm

Intensity (mW)
Intensity (mW)

Intensity (mW)
6 6 6.0 1536.168 nm
Intensity (V)

10.10 dB
6.85 dB
1563.82 nm
1592.472 nm
1540.308 nm
2 4.5 1567.976 nm
1596.62 nm
1544.652 nm
1572.356 nm
3 3 3.0 1601.028 nm
1549.096 nm
1 1576.748 nm
1605.05 nm
1.5 1553.436 nm
1581.132 nm
1609.932 nm

0 0 0 0.0
6150 6230 6310 6390 6470 6550 6630

00

00

00
00

00

00
0

00

00

00

00

00
20

40

60

80

58

63

68
58

63

68
10

12

14

16

18
–1
Wavenumber (cm )
Electric power (mW) Wavenumber Wavenumber
(cm–1) (cm–1)

Fig. 5 Single wavelength characterization. a Detected output interferogram with TLS-1 input (set at 1584.620 nm) with 6 mW input power when resonance
wavelength shift Δλ = 3.38 nm. b Retrieved spectra with TLS-1 input at 8 mW input power when Δλ = 3.38 nm. The on-resonance wavelength λon3 =
1584.620 nm and the value of dλ is set to be 0.3 and 0.47 nm, respectively. On-resonance wavelengths are denoted in black and off-resonance
wavelengths in red. c Retrieved spectra with TLS-1 input at on-resonance wavelengths when Δλ = 3.38, 7.38, 11.38, 15.38, 19.38, 23.38, and 27.38 nm,
respectively

a b Retrieved
c 1.2
on2 : 3 mW 0.92
4.8 Source
on3 : 6 mW 0.69 Retrieved
0.46 1.0 Source

Normalized power (a.u.)

3.2 d = 1.880 nm
Normalized power (a.u.)

0.23
0.00 0.8
Intensity (mW)

1.6 0.92 Retrieved

Source
0.69
0.0 0.6
0.46
on1 : 6 mW
0.23
4.8 on2 : 3 mW d = 0.448 nm
0.00 0.4
0.92 Retrieved
3.2 Source
0.69
0.2
1.6 0.46
0.23 d = 0.476 nm
0.0 0.00 0.0
00

00

00

00

00

00

00

32

44

56

68

80

92

04

16
32

33

34

35

36

37

38
50

55

60

65

70

75

80

15

15

15

15

15

15

16

16
15

15

15

15

15

15

15

Wavenumber (cm–1) Wavelength (nm) Wavelength (nm)

Fig. 6 Double wavelengths and broadband spectrum characterization. a Retrieved spectra with TLS-1 and TLS-2 (set at adjacent on-resonance wavelengths
of MRR, respectively) input simultaneously when Δλ = 0. b Normalized retrieved spectra (black) using the spectra with two broad spectral peaks input
(red). c Normalized retrieved spectrum (black) with a broadband source (red) input. The source spectrum is generated from an optical fiber interferometer

on-resonance wavelengths of the MRR when Δλ = 3.38, 7.38, criterion of the tunable MZI (19.32 nm). Furthermore, we per-
11.38, 15.38, 19.38, 23.38, and 27.38 nm, respectively. It is shown formed a broadband signal measurement with minimum MRR
that the tunable MZI can retrieve each filtered spectrum from the tuning value of 0.47 nm. The transmission spectrum from an
drop port of the MRR with single wavelength input by thermal optical fiber interferometer is used as the input. For broadband
tuning within one FSR of 28 nm. signal input, all the detected input sparse spectra are retrieved
using the normalization coefficient matrix A (see Supplementary
Fig. 9) and are then combined to produce the original input
Double wavelength characterization. TLS-1 and TLS-2 (ANDO
spectrum. The normalized retrieved spectrum and input broad-
AQ4321D) are combined with a 50/50 optical coupler as the
band source are shown in Fig. 6c. The retrieved spectrum agrees
input for double wavelength characterization. Figure 6a shows the
well with the input spectrum. The small discrepancy is due to
retrieved spectra when TLS-1 and TLS-2 are set at adjacent on-
misalignment between lensed fiber and inverse-taper waveguide
resonant wavelengths. The tuning state is resonance wavelength
coupler while heating MZI. Another reason is resonance position
shift Δλ = 0. The resonance wavelengths are λon1 = 1525.400 nm,
fluctuation due to thermal crosstalk (see Supplementary Note 3).
λon2 = 1552.844 nm, and λon3 = 1581.240 nm, respectively. One
By packaging the lensed fiber to the input waveguide, the mis-
can see that the two adjacent on-resonant wavelength compo-
alignment would not present. The thermal crosstalk mainly ori-
nents can be easily distinguished and reconstructed by the tunable
ginates from silicon substrate since the buried oxide layer (BOX)
MZI. As a result, the tunable MZI can retrieve each filtered
is not thick enough to effectively isolate the heat from the heater
spectrum at each tuning state of the MRR.
above the MZI to Si substrate (as in our experiment, the BOX is
2 μm). By employing isolation trenches around MRR and thermal
Broadband spectrum recovery. To further test the resolution, we compensation (see Supplementary Note 3), the stability tolerance
use a wavelength-division multiplexer (Sharetop WDM) to gen- of the resonance wavelength has been decreased from 2δλ
erate two broad spectral peaks as the input spectrum of the RAFT (without thermal compensation) to δλ/5. The current value can
spectrometer27,33. The normalized retrieved spectra and the input be further decreased by reducing residue thermal crosstalk,
spectra are shown in Fig. 6b. It can be seen that the minimum optimized thermal compensation and/or adopting heater with
resolvable wavelength detuning is 0.448 nm, which is smaller than low-temperature coefficient of resistance. The residue thermal
the minimum MRR tuning value of 0.47 nm. Hence, the resolu- crosstalk can be further mitigated through fabricating isolation
tion is 0.47 nm, which significantly outperforms the Rayleigh trenches near MZI arms and can also be effectively reduced by

6 NATURE COMMUNICATIONS | (2019)10:2349 | https://doi.org/10.1038/s41467-019-10282-1 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10282-1 ARTICLE

making both MRR and MZI fully suspended34 (see Supplemen- non-linearity and thermal expansion effect. Moreover, the longer
tary Note 3) and/or using thicker BOX. Moreover, it will be well waveguide length will induce larger footprint and higher optical
compensated with a feedback circuit to control the applied power loss due to imperfect fabrication. The higher optical loss will in
on MRR and MZI heater. turn reduce the contrast of interferogram, leading to reduced
SNR. It is very challenging to achieve a resolution down to
sub-nm using a tunable MZI. It is suggested that the resolution of
Discussion the tunable MZI, R ≥ 10 nm, when moderate power consumption
The designed RAFT spectrometer consisting of a tunable MRR and and arm length are required and resolution enhancement (R/δλ)
a tunable MZI enhances the resolution dramatically far beyond the is larger than 20 times. Hence, m is chosen as m ≤ 9. Time-
Rayleigh criterion of a typical tunable MZI (42.6-fold here). Due to multiplexing will also induce extra power consumption due to
the employment of MRR filter, the MZI only needs a resolution multiple scans of MRR and MZI. Nearly, 2% (δλ/FSR) of total
(<28 nm) to resolve the resonance wavelengths with a mini- time is spent for measuring a single resolution element and in our
mum span of one FSR of the MRR, significantly easing the current experiment, the one-time scan duration is 2 s. The time-
requirement on the maximal OPD. Since thermal isolation trenches scale measurement on thermal response time of MZI and MRR is
are employed for the MRR, the power consumption is significantly presented in Supplementary Note 4. The results show that the
reduced. The total energy consumed by MRR and MZI for N scans maximum sweeping frequency of MZI is 10 kHz. For 10 kHz
are 1.96 and 67.2 J, respectively. The calculations are presented in sampling frequency and 2000 one-time sampling points, the one-
Supplementary Note 3. The power consumption of MZI can be time scan duration is reduced to 0.2 s and the total time is
reduced to 150 mW through fabrication of isolation trenches along reduced to 0.2FSR/δλ ≈11.4 s. Hence, the fast sweeping frequency
the waveguides of MZI arms (see Supplementary Note 3). Hence, of MZI will compensate the gain loss due to time-multiplexing
the total energy consumed by MZI can be reduced from 67.2 to 5.6 and reduce the total energy consumption of MZI and MRR.
J. Note that the Si substrate under the MRR in the tested RAFT In conclusion, a microring RAFT spectrometer is experimen-
spectrometer and the MZI arm in the testing structure is not totally tally demonstrated with a tunable MRR, a tunable MZI, and a Ge-
removed (Supplementary Fig. 11b), the heating efficiency of both on-SOI PD being integrated onto a single chip. The tunable MRR
MRR and MZI can be further improved (~8.75 times) if the pre-filters the input spectrum into a sparse spectrum to match the
waveguides are fully suspended35. The resolution limit of the tun- resolution of the following cascaded tunable MZI. Due to the
able MZI due to waveguide dispersion is 18 ≤ R ≤ 19.9 nm in the high-quality factor (~9661) of the MRR, the resolution of the
detected wavelength range with fixed maximal OPD (with maximal RAFT spectrometer is dramatically boosted far beyond the Ray-
heating power of 1.8 W employed in the experiment). In our pro- leigh criterion of a typical FT spectrometer by finely tuning the
posed structure, the final resolution δλ can be further improved by resonance wavelength of MRR. A high-resolution of 0.47 nm and
increasing the Q value through coupler design (e.g., optimizing the a large bandwidth of ~90 nm is achieved. The bandwidth can be
gap and/or coupling length of the coupling region) and decreasing largely extended by integrating a paralleled RAFT element array.
the losses in the ring waveguide and couplers via fabrication opti- The power consumption due to thermal tuning and time-
mization. For instance, for single-pass amplitude transmission a = multiplexing can be drastically reduced by introducing isolation
0.9986, if Q ≥ 10,000, i.e., δλ ≤ 0.153 nm at 1528.256 nm, the self- trenches and increasing the sweeping frequency of MZI. The SNR
coupling coefficient r ≥ 0.9835 (see Supplementary Note 5). The degraded by time-multiplexing can be improved by reducing
parameter values for this calculation are shown in Supplementary optical loss and/or adopting smaller FSR. It has high potential for
Table 3. Hence, the gap between ring and straight waveguide is applications such as chemical and biological sensing, on-chip
larger than 230 nm according to FDTD simulation results. Noticing spectroscopy, and image spectrometry.
that the gap dominates in determining r, thereby, the Q value, only
the fabrication tolerance of gap is considered here. Since the
transmitted power from MRR will be reduced when increasing Q Methods
Fabrication. The microring RAFT spectrometer is fabricated from an 8-inch SOI
value (see Supplementary Fig. 16b), the designed gap is 240 nm with wafer using the nano-silicon photonic fabrication technology. After fabricating the
20 nm tolerance, i.e., ±10 nm fabrication deviation, which can be Si waveguides structures, several implantation processes and Ge epitaxy are done
easily achieved by the current fabrication technology (±7.5 nm for fabrication of the waveguide-coupled PD. Subsequently, a 1 μm-thick upper
deviation). Although the working spectral window depends on the silicon dioxide (SiO2) cladding layer is deposited and then a thin layer of titanium
nitride (TiN) is formed to act as the resistive layer for heaters. Subsequently, an
transmission band of various components such as waveguides, aluminum (Al) thin film is patterned for electrical connection to power the heaters
couplers, beam splitters, and PD, etc., the bandwidth can be dras- and PD. At last, the isolation trenches are formed through etching SiO2 cladding
tically extended by designing a paralleled RAFT spectrometer array. and Si substrate.
The MRR before the MZI will compromise the Fellgett
advantage of a typical FT spectrometer, which will induce a lower Experiment setup. The experiment setup for MRR characterization is shown in
SNR. The calculations are presented in Supplementary Note 4. Supplementary Fig. 17a. A broadband light source (Amonics C + L Band ASE
Hence, the SNR requirement needs to be considered, since it will broadband light source) is used to perform MRR characterization. It is firstly
limit the minimum resolution value as shown in Supplementary coupled to a polarization beam splitter and a polarization controller to ensure that
Fig. 16c with simulation parameter values summarized in Sup- only TE-polarized light is input into the on-chip spectrometer. The device under
test, i.e., the on-chip spectrometer chip with a thermo electric coolor as the sub-
plementary Table 3. The multiplex gain loss is approximately strate to control and stabilize the temperature using a temperature controller, is
87.5% with m = 3 in our experiment. With decreased SNR, the mounted on the XYZ stage holder for fiber-chip alignment. The output spectrum
level of the minimum detectable signal is increased, thus leading from the throughput port of the MRR is detected by an optical spectrum analyzer
to reduced dynamic range. This loss can be reduced by appro- (OSA, Yokogawa AQ6370D). The applied voltages of heater 1 and heater 2 are
controlled by a laptop via a microcontroller.
priately increasing m, i.e., employing an MRR with larger cir- The experiment setup for spectrometer characterization (single and double
cumference which has smaller FSR. At the same time, the wavelength characterization and broadband spectrum recovery) is shown in
resolution R (equal to FSR) of the tunable MZI must be improved Supplementary Fig. 17b. TLS-1 (santec TSL-510) is adopted as the input only
accordingly (see Supplementary Fig. 14). To improve resolution, a for single wavelength characterization. TSL-1 and TLS-2 (ANDO AQ4321D) are
used simultaneously to perform double wavelength characterization. Light from
larger maximal OPD Δ, i.e., more heating power and/or longer TLS-1 and TLS-2 is combined with a 50/50 optical coupler as the input. The output
arm length are required. The increased heating power not only light from MZI is detected by an off-chip photodetector (PD, Thorlabs PDA-10CS-
increases the power consumption, but also brings larger TO EC) of which the intensity signal is acquired by a laptop via a microcontroller.

NATURE COMMUNICATIONS | (2019)10:2349 | https://doi.org/10.1038/s41467-019-10282-1 | www.nature.com/naturecommunications 7

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-10282-1

Electrical measurements. The spectrometer chip is wire-bonded to a PCB board 22. Podmore, H. et al. Demonstration of a compressive-sensing Fourier-transform
for electric connections with an external PCB board which is subsequently con- on-chip spectrometer. Opt. Lett. 42, 1440–1443 (2017).
nected to a microcontroller. The external PCB board is integrated with DACs and 23. Podmore, H. et al. Athermal planar-waveguide Fourier-transform
an operational power amplifier (OPA) to amplify the electrical signal for MZI spectrometer for methane detection. Opt. Express 25, 33018–33028 (2017).
heater. The resistances of MRR heater and MZI heater are ~0.547 and ~1.277 kΩ, 24. Nedeljkovic, M. et al. Mid-infrared silicon-on-insulator Fourier-transform
respectively. spectrometer chip. IEEE Photon. Tech. Lett. 28, 528–531 (2016).
25. Herrero-Bermello, A. et al. Temperature dependence mitigation in stationary
Thermal compensation. Even though the isolation trenches fabricated near MRR Fourier-transform on-chip spectrometers. Opt. Lett. 42, 2239–2242 (2017).
have effectively reduced the thermal crosstalk from MZI heater, there is still a 26. Florjańczyk, M. et al. Development of a slab waveguide spatial heterodyne
remaining resonance wavelength shift (maximal 1 nm) due to thermal crosstalk. spectrometer for remote sensing. SPIE MOEMS-MEMS 7594, 9 (2010).
Hence, thermal compensation is performed to stabilize the resonance position. The 27. Bock, P. J. et al. Subwavelength grating Fourier‐transform interferometer array
thermal compensation method is presented in Supplementary Note 3. in silicon-on-insulator. Laser Photonics Rev. 7, L67–L70 (2013).
28. Zheng, S., Cai, H., Gu, Y. D., Chin, L. K. & Liu, A. Q. On-chip Fourier
transform spectrometer for chemical sensing applications. Conf. Lasers Electro
Broadband source calibration. The description of the calibration process is pre-
Opt. 2, AM1J.6 (2016).
sented in Supplementary Note 2.
29. Souza, M. C., Grieco, A., Frateschi, N. C. & Fainman, Y. Fourier transform
spectrometer on silicon with thermo-optic non-linearity and dispersion
Data availability correction. Nat. Commun. 9, 665 (2018).
The data that support the findings of this study are available from the corresponding 30. Rabus, D. G. Integrated Ring Resonators: The Compendium. (Springer, Berlin,
authors on reasonable request. 2007).
31. Economou, E. N. Green’s Functions in Quantum Physics. (Springer, Berlin, 1983).
32. Griffiths, P. R., & De Haseth, J. A. Fourier Transform Infrared Spectrometry.
Received: 27 July 2018 Accepted: 30 April 2019 (John Wiley & Sons, Hoboken, 2007).
33. Kita, D. M. et al. High-performance and scalable on-chip digital Fourier
transform spectroscopy. Nat. Commun. 9, 4405 (2018).
34. Padmaraju, K. & Bergman, K. Resolving the thermal challenges for silicon
microring resonator devices. Nanophotonics 3, 269–281 (2014).
35. Dong, P. et al. Thermally tunable silicon racetrack resonators with ultralow
References tuning power. Opt. Express 18, 20298–20304 (2010).
1. Venghaus, H., & Grote, N. Fibre Optic Communication: Key Devices.
Chapter 14 (Springer, Cham, 2017).
2. Lockwood, D. & Pavesi, L. Silicon Photonics II: Components and Integration Acknowledgements
Series: Topics in Applied Physics. (Springer, Berlin, 2011). This work was supported by Ministry of Education, Singapore under Tier 3 (MOE2017-
3. Stojanović, V. et al. Monolithic silicon-photonic platforms in state-of-the-art T3-1-001) and Tier 1 (RG182/16 (S)) and the Singapore National Research Foundation
CMOS SOI processes [Invited]. Opt. Express 26, 13106–13121 (2018). under the Competitive Research Program (NRF-CRP13-2014-01) and the Incentive for
4. Ryckeboer, E., et al. Spectroscopic sensing and applications in silicon photonics. Research & Innovation Scheme (1102-IRIS-05-04) administered by PUB. The authors
In Proceedings of the IEEE 14th International Conference 81–82 (2017). acknowledge Liu Yong Sheng for kind help and fruitful discussions.
5. Shi, Y. Z. et al. Nanometer-precision linear sorting with synchronized
optofluidic dual barriers. Sci. Adv. 4, eaao0773 (2018).
6. Shi, Y. Z. et al. Sculpting nanoparticle dynamics for single-bacteria-level screening Author contribution
and direct binding-efficiency measurement. Nat. Commun. 9, 815 (2018). S.N.Z. and A.Q.L. jointly conceived the idea. S.N.Z., J.Z., and J.F.S. performed the
7. Ryckeboer, E. et al. Silicon-on-insulator spectrometers with integrated numerical simulations and theoretical analysis. S.N.Z., H.C., and D.L.K. did the fabri-
GaInAsSb photodiodes for wide-band spectroscopy from 1510 to 2300 nm. cation of the device. S.N.Z. and J.Z. did experiments of spectrometer characterization. S.
Opt. Express 21, 6101–6108 (2013). N.Z., J.Z., L.K.C., P.Y.L., Z.P.L., and A.Q.L. prepared the paper. J.F.S., Z.P.L., and A.Q.L.
8. Cheben, P. et al. A high-resolution silicon-on-insulator arrayed waveguide supervised and coordinated all the work. All authors commented on the paper.
grating microspectrometer with sub-micrometer aperture waveguides. Opt.
Express 15, 2299–2306 (2007).
9. Muneeb, M. et al. III-V-on-silicon integrated micro-spectrometer for the 3 μm
Additional information
Supplementary Information accompanies this paper at https://doi.org/10.1038/s41467-
wavelength range. Opt. Express 24, 9465–9472 (2016).
019-10282-1.
10. Muneeb, M. et al. Demonstration of Silicon-on-insulator mid-infrared
spectrometers operating at 3.8 μm. Opt. Express 21, 11659–11669 (2013).
Competing interests: The authors declare no competing interests.
11. Malik, A. et al. Germanium-on-silicon planar concave grating wavelength (de)
multiplexers in the mid-infrared. Appl. Phys. Lett. 103, 161119 (2013).
Reprints and permission information is available online at http://npg.nature.com/
12. Kyotoku, B. B., Chen, L. & Lipson, M. Sub-nm resolution cavity enhanced
reprintsandpermissions/
micro-spectrometer. Opt. Express 18, 102–107 (2010).
13. Momeni, B., Hosseini, E. S., Askari, M., Soltani, M. & Adibi, A. Integrated
Journal peer review information: Nature Communications thanks Amir Khodabakhsh
photonic crystal spectrometers for sensing applications. Opt. Commun. 282,
and other anonymous reviewer(s) for their contribution to the peer review of this work.
3168–3171 (2009).
Peer reviewer reports are available.
14. Momeni, B., Hosseini, E. S. & Adibi, A. Planar photonic crystal
microspectrometers in silicon-nitride for the visible range. Opt. Express 17,
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in
17060–17069 (2009).
published maps and institutional affiliations.
15. Redding, B., Liew, S. F., Sarma, R. & Cao, H. Compact spectrometer based on
a disordered photonic chip. Nat. Photon. 7, 746–751 (2013).
16. Xia, Z. et al. High resolution on-chip spectroscopy based on miniaturized
microdonut resonators. Opt. Express 19, 12356–12364 (2011). Open Access This article is licensed under a Creative Commons
17. Redding, B., Liew, S. F., Bromberg, Y., Sarma, R. & Cao, H. Evanescently Attribution 4.0 International License, which permits use, sharing,
coupled multimode spiral spectrometer. Optica 3, 956–962 (2016). adaptation, distribution and reproduction in any medium or format, as long as you give
18. Yu, K., Lee, D., Krishnamoorthy, U., Park, N. & Solgaard, O. Micromachined appropriate credit to the original author(s) and the source, provide a link to the Creative
Fourier transform spectrometer on silicon optical bench platform. Sens. Commons license, and indicate if changes were made. The images or other third party
Actuat. A Phys. 130, 523–530 (2006). material in this article are included in the article’s Creative Commons license, unless
19. Manzardo, O., Herzig, H. P., Marxer, C. R. & de Rooij, N. F. Miniaturized indicated otherwise in a credit line to the material. If material is not included in the
time-scanning Fourier transform spectrometer based on silicon technology. article’s Creative Commons license and your intended use is not permitted by statutory
Opt. Lett. 24, 1705–1707 (1999). regulation or exceeds the permitted use, you will need to obtain permission directly from
20. Nie, X., Ryckeboer, E., Roelkens, G. & Baets, R. CMOS-compatible broadband the copyright holder. To view a copy of this license, visit http://creativecommons.org/
co-propagative stationary Fourier transform spectrometer integrated on a licenses/by/4.0/.
silicon nitride photonics platform. Opt. Express 25, A409–A418 (2017).
21. Le Coarer, E. et al. Wavelength-scale stationary-wave integrated Fourier-
transform spectrometry. Nat. Photon. 1, 473 (2007). © The Author(s) 2019

8 NATURE COMMUNICATIONS | (2019)10:2349 | https://doi.org/10.1038/s41467-019-10282-1 | www.nature.com/naturecommunications