Fabrication-tolerant Fourier transform
spectrometer on silicon with broad bandwidth
and high resolution
ANG LI,
1,
*JORDAN DAVIS,
1
ANDREW GRIECO,
1
NAIF ALSHAMRANI,
1,2
AND YESHAIAHU FAINMAN
1,3
1
Department of Electrical and Computer Engineering, University of California, San Diego, California 92093, USA
2
King Abdulaziz City for Science and Technology (KACST), Riyadh 11442, Saudi Arabia
3
e-mail: fainman@ece.ucsd.edu
*Corresponding author: angli@ucsd.edu
Received 2 October 2019; revised 9 December 2019; accepted 16 December 2019; posted 18 December 2019 (Doc. ID 379184);
published 31 January 2020
We report an advanced Fourier tran sform spectrometer (FTS) on silicon with significant improvement compared
with our previous demonstration in [Nat. Commun. 9, 665 (2018)]. We retrieve a broad band spectrum (7 THz
around 193 THz) with 0.11 THz or sub nm resolution, more than 3 times higher than previously demonstrated
[Nat. Commun. 9, 665 (2018)]. Moreover, it effectively solves the issue of fabrication variation in waveguide
width, which is a common issue in silicon photonics. The structure is a balanced Mach–Zehnder interferomet er
with 10 cm long serpentine waveguides. Quasi-continuous optical path difference between the two arms is in-
duced by changing the effective index of one arm using an integrated heater. The serpentine arms utilize wide
multi-mode waveguides at the straight sections to reduce propagation loss and narrow single-mode waveguides at
the bending sections to keep the footprint compact and avoid modal crosstalk. The reduction of propagation loss
leads to higher spectral efficiency, larger dynamic range, and better signal-to-noise ratio. Also, for the first time to
our knowledge, we perform a thorough systematic analysis on how the fabrication variation on the waveguide
widths can affect its performance. Additionally, we demonstrate that using wide waveguides efficiently leads to a
fabrication-tolerant device. This work could further pave the way towards a mature silicon-based FTS operating
with both broad bandwidth (over 60 nm) and high resolution suitable for integration with various mobile
platforms.
© 2020 Chinese Laser Press
https://doi.org/10.1364/PRJ.379184
1. INTRODUCTION
Fourier transform spectrometers (FTSs) recovering an un-
known spectrum are powerful tools in various fields including
chemical sensing, bio-medical applications, and autonomous
vehicles, etc. [1–4]. Compared to other direct-detection-based
spectrometers, such as monochromators or grating-based spec-
trum analyzers, one of the key advantages of FTSs is Fellgett’s
advantage, namely, higher signal-to-noise ratio and dynamic
range [5]. The FTS employs an interferogram (i.e., autocorre-
lation function) generated from the input signal with unknown
spectra using an interferometer, where the input signal is di-
vided into two paths with variable optical path difference
(OPD). At the output of the interferometer, the optical signals
from these two optical paths are recombined on a photodetec-
tor, which is used to generate a photocurrent as a function
of OPD, thereby generating the desired interferogram. The
Fourier transform of the detected interferogram (i.e., autocor-
relation function) produces the power spectral density of the
input signal. The larger the maximum OPD, the higher the
spectral resolution. A classical FTS typically uses a moving
element (such as a mirror) to introduce quasi-continuous
tunable OPD between two paths and generates a temporal in-
terferogram (i.e., autocorrelation function). Thus, FTSs are
usually realized as bulky free-space interferometric systems
and suffer from slow scanning speeds, resulting in long mea-
surement times.
Use of silicon photonics as a platform for FTS realization has
attracted widespread attention as a candidate to meet the rap-
idly growing demand for low-cost, portable devices [6]. In con-
trast to free-space FTSs, most FTSs demonstrated on silicon
platforms are spatial heterodyne spectrometers (SHSs) that em-
ploy a large array of unbalanced Mach–Zehnder interferome-
ters (MZIs) with varying length differences, each of which
introduces a fixed amount of OPD to the spectrum [3,7–9].
In order to achieve high resolution (i.e., large OPD), the inci-
dent spectral signal is typically required to be split into over
Research Article
Vol. 8, No. 2 / February 2020 / Photonics Research 219
2327-9125/20/020219-06 Journal © 2020 Chinese Laser Press