RES EARCH

◥ lengths, such that analysis of complete spectra

REVIEW SUMMARY can be achieved either with a single filter (the
transmissive properties of which can be varied
NANOPHOTONICS over time) or by passing light through an array

Miniaturization of optical spectrometers

of multiple unique narrowband filters, each
mounted onto its own detector; (iii) Fourier
transform systems, where integrated interfero-
Zongyin Yang*, Tom Albrow-Owen*, Weiwei Cai†, Tawfique Hasan† meters [such as those based on microelec-
tromechanical systems (MEMS) components]
can be used to produce temporal or spatial
BACKGROUND: Optical spectrometry is one of Toward this aim, an approach that involves interferograms, which are then computation-
the most powerful and widely used charac- simply scaling down benchtop systems (with ally converted to a readable spectrum; and (iv)
terization tools in scientific and industrial miniaturized gratings and reflective optics) a newly emerging paradigm of microspec-
research. Benchtop laboratory spectrometer becomes constrained as a result of the com- trometers, in which computational techniques
systems—characterized by bulky optical compo- plex fabrication involved and the inherent are used to approximate or reconstruct an in-
nents, moving parts, and long path lengths— proportionality of resolution to path length cident light spectrum from precalibrated spec-
can deliver unparalleled, ultrafine resolution in dispersion-based systems. tral response information encoded within a set
and wide spectral ranges. However, a rapidly of broadband detectors or filters.
growing application space exists for spectral ADVANCES: A wide variety of miniaturized
analysis where the need for reduced physical spectrometer systems have emerged since OUTLOOK: We now stand at a watershed where
dimensions, cost, or power consumption takes the early 1990s. These can be grouped into this field is yielding ultracompact micro-
precedence over the need for high perform- four broad categories according to the under- spectrometer systems with performance and

Downloaded from https://www.science.org on December 29, 2023

ance. The demand for portable or handheld lying strategies they use for spectral charac- footprint near those viable for integrated appli-
spectral analysis devices requires shrinking of terization: (i) those that have tried to push the cations such as lab-on-a-chip systems, smart-
these systems down to centimeter-scale foot- boundaries of miniaturization using a conven- phones, and spectral imagers. Until recently,
prints. More extreme miniaturization to sub- tional benchtop strategy, where light interacts advancement has been inspired by and has ben-
millimeter length scales would open a range of with miniaturized dispersive optics such that efited from wider technological trends in the
opportunities for in situ analysis, with poten- different spectral components are spatially production of hardware. For instance, earlier
tial for integration into lab-on-a-chip systems, separated when arriving at a detector array; dispersion-based strategies have been improved
smartphones, or even spectrometer-per-pixel (ii) narrowband filters, which can be used to through optimization of high-precision micro-
snapshot hyperspectral imaging devices. selectively transmit light with specific wave- fabrication, lithographic, and etching tech-
niques to produce ever more scaled-down
gratings and optics. In parallel, the devel-
opment of MEMS components has enabled
ics Na
pt rro ultracompact, electronically driven moving
o wb parts for miniaturized Fourier transform
ve
si
an

interferometer–based devices. However, as the

er

d fi
sp

physical size and cost of processing power

lte
Di

have fallen sharply over the past 15 years, the

rs

emergence of reconstructive microspectrome-

ters has heralded a fundamental shift in the
field, where developments in the software will
shoulder much of the burden for enhancing
device performance while footprints continue
to shrink. Maturation of the algorithmic strat-
egies behind these devices will likely see the
incorporation of machine learning–based
techniques, which increasingly will be able to
compensate for the compromises in detector
performance necessitated by further minia-
Tunable
turization. This represents a promising route
ΔL or Δn toward ultracompact high-performance sys-
tems and the emergence of spectral analysis
m
for
Re

in a host of previously inaccessible platforms

ns
c

in scientific research, industry, and consumer

▪
on

tr a
st

electronics.
ru

er

tiv ri
c

e u
Fo The list of author affiliations is available in the full article online.
*These authors contributed equally to this work.
†Corresponding author. Email: [email protected] (W.C.);
[email protected] (T.H.)
Cite this article as Z. Yang et al., Science 371, eabe0722
(2021). DOI: 10.1126/science.abe0722
Strategies toward ultracompact microspectrometers. Schemes for miniaturized spectral sensing systems
based on dispersive optics, narrowband filters, Fourier transform interferometers, and computational spectral READ THE FULL ARTICLE AT
reconstruction schemes have all emerged over the past three decades. https://doi.org/10.1126/science.abe0722

Yang et al., Science 371, 480 (2021) 29 January 2021 1 of 1

RES EARCH

◥ (Fig. 1C). Early miniaturized spectrometers

REVIEW fell within these first three classifications
and featured designs that largely resembled
NANOPHOTONICS scaled-down benchtop spectrometers, with

Miniaturization of optical spectrometers

out-of-plane diffractive optics or microelec-
tromechanical systems (MEMS)–based inter-
ferometers. Further development saw such
Zongyin Yang1,2*, Tom Albrow-Owen1*, Weiwei Cai3†, Tawfique Hasan1,4† designs largely give way to planar systems
based on waveguides and integrated optics
Spectroscopic analysis is one of the most widely used analytical tools in scientific research and (14–19).
industry. Although laboratory benchtop spectrometer systems offer superlative resolution and However, in the past decade, a fourth cat-
spectral range, their miniaturization is crucial for applications where portability is paramount or egory has emerged as a new paradigm of mi-
where in situ measurements must be made. Advancement in this field over the past three decades is crospectrometer devices. “Reconstructive” or
now yielding microspectrometers with performance and footprint near those viable for lab-on-a-chip “computational” spectrometer systems (Fig.
systems, smartphones, and other consumer technologies. We summarize the technologies that 1D) take advantage of more readily available
have emerged toward achieving these aims—including miniaturized dispersive optics, narrowband computer processing power and reductions in
filter systems, Fourier transform interferometers, and reconstructive microspectrometers—and microprocessor size and cost. They typically
discuss the challenges associated with improving spectral resolution while device dimensions shrink feature a set of detectors encoded with distinc-
ever further. tive spectral response characteristics, which,
when measured in parallel, can be combined

O
using complex algorithms to approximate or
ptical spectrometers have served as the other hand, in industry, devices suitable “reconstruct” an incident light spectrum. Such

Downloaded from https://www.science.org on December 29, 2023

one of the most important instruments for drone-based spectral imaging—that is, systems can harness not only technological
to date for materials characterization where spectral information is correlated with advances in hardware, but also the develop-
and chemical analysis (1). Conventional spatial data—could revolutionize large-scale ment of new computational approaches—in
benchtop spectrometers typically rely crop monitoring. particular, those based on compressive sensing
on a combination of bulky dispersive optics, In general, reducing the size of a spectrom- and machine learning.
long optical path lengths, detector arrays, and eter necessitates a compromise with respect to
movable parts. These requirements impede degradation of its resolution, dynamic range, Miniaturized dispersive optics
their miniaturization for applications where or signal-to-noise ratio. However, microspec- Conventional spectrometers typically con-
it is critical to minimize size, cost, and power trometers can be engineered to meet “ac- sist of one or several diffraction gratings, an
consumption. Recent years have seen the de- ceptable” levels of performance for specific optical path, and a detector array. Light passes
velopment of scaled-down spectrometer sys- applications (7, 10). In many cases, where the through an input slit and is collimated onto
tems for a wide range of handheld, portable, goal is identification of signature spectral a diffraction grating that disperses spectral
and integrated applications, including soil and peaks rather than relative metrology, a “satis- components in different directions. A concave
crop analysis, monitoring of food industry factory” resolution in the visible range may, mirror focuses this dispersed light toward the
production lines, and marine/underwater for instance, be on the order of 10 nm, or even detector array. Advancements in micro- and
scientific research (2–6). It is clear that in these larger (5). Moreover, by enhancing a particular nanofabrication techniques have provided an
uses, it is often far preferable to attain indica- aspect of performance, microspectrometers opportunity to develop microspectrometers
tive, instantaneous, on-the-spot results, rather can be specialized for extreme measurements by scaling down the components of these sys-
than transporting samples to a laboratory for that are challenging to implement using a tems. A wide range of increasingly miniatur-
ultrahigh-resolution analysis (5, 7). Further conventional system. For example, the relative ized, spatially dispersive spectrometers with
miniaturization, down to the submillimeter strengths of on-chip single-photon spectrom- centimeter-scale footprints have been demon-
scale, could provide opportunities in a wide eters (11, 12) and single-nanowire spectrometers strated since the 1990s (20, 21). These disper-
range of applications, including lab-on-a-chip (13) lie in their ultrahigh-sensitivity detection sive microspectrometers have typically been
spectroscopy and other in situ or even in vitro and ultracompact footprint, respectively. fabricated via wafer bonding, with optical paths
characterization systems. Various possibilities Since the early 1990s, miniaturized optical fashioned through electrochemically controlled
can be envisaged for consumer technologies, spectrometers based on a wide variety of de- etching (Fig. 2A).
such as in smartphone-based devices (8, 9), for signs and working principles have been dem- As the system footprint and thus the com-
applications including the detection of coun- onstrated, with a range of operational spectral ponent size decreases, a number of factors
terfeit pharmaceuticals and banknotes, moni- bands and resolutions. Below, we summarize must be considered. Separation of spectral
toring of skin health, or even determining the the most explored technological platforms, pre- components at the detector plane depends on
sugar and fat content in food products. On senting their relative merits and drawbacks. the distance light is allowed to travel after
We have broadly organized the field into four meeting the dispersive element. As such, for
1
main categories, representing the most prom- a given grating and detector array, the reso-
Department of Engineering, University of Cambridge,
Cambridge CB3 0FA, UK. 2College of Information Science
inent strategies for identifying different spec- lution Dl [the full width at half maximum
and Electronic Engineering, State Key Laboratory of Modern tral components. The first three categories are (FWHM) of the narrowest distinguishable
Optical Instrumentation, Zhejiang University, Hangzhou (i) those that feature dispersive optics to split spectral component, in nanometers] is pro-
310027, China. 3Key Laboratory of Education Ministry for
light toward spatially separated detectors (Fig. portional to the optical path length of the sys-
Power Machinery and Engineering, School of Mechanical
Engineering, Shanghai Jiao Tong University, Shanghai 1A), (ii) those that use narrowband filters to tem. When the device is made more compact,
200240, China. 4Cambridge Graphene Centre, University of preferentially transmit particular spectral com- the path length also necessarily decreases, thus
Cambridge, Cambridge CB3 0FA, UK. ponents to different detectors (Fig. 1B), and (iii) lowering the spectral resolution. This can be
*These authors contributed equally to this work.
†Corresponding author. Email: [email protected] (W.C.); Fourier transform microspectrometers based compensated to an extent by increasing the
[email protected] (T.H.) around temporal or spatial interferometers detector’s pixel density within a given width.

Yang et al., Science 371, eabe0722 (2021) 29 January 2021 1 of 11

RES EARCH | R E V I E W

A Dispersive optics B Narrowband filters

Detector array d
ΔL

Detector
Reflection diffraction Transmission diffraction Waveguide dispersion Tunable filter Tunable filter Filter array Variable filter
(in / out plane) (in / out plane) Δd Δ

C Fourier transform D Reconstructive

Mirror
ΔL1
50:50 Patterns at different λ Spectral transmissions Spectral responses
Light Splitter ΔL2 λ2 λ2 λ2
Tunable .
p ( ,x) s( ) d = I(x) D ( ) .s( ) d
i
= Ii .
Di( ) s( ) d = Ii
ΔL or Δn λ1 λ1 λ1
...

Mirror
s(λ) Reconstructed

Downloaded from https://www.science.org on December 29, 2023

s(λ) Reconstructed
MEMS ΔLn s(λ) Reconstructed
Detector scanning * *
mirror Target pattern
*
Michelson Mach-Zehnder Mach-Zehnder Stationary wave * Dispersive media * Broadband * Broadband
interferometer interferometer interferometer array integrated filter array detector array

Fig. 1. Different underlying strategies for miniaturized spectrometer systems that have emerged over the past 30 years. (A) Miniaturized dispersive optics.
(B) Tunable or arrayed narrowband filters. (C) Fourier transform–based systems. (D) Computational spectral reconstruction–based systems.

However, the quality of optical components of the substrate, followed by deposition of a confinement, advantages present themselves
and the system alignment present conflicting waveguide layer on top. As illustrated in Fig. in their straightforward integration into mo-
problems in fabrication; for instance, etching- 2D, in these systems, light is coupled into the nolithic, waveguide-based optical analysis
induced surface roughness will cause more waveguide via an input grating. When propa- systems.
light to be scattered before arriving at the gating through the waveguide, light interacts In addition to innovations with respect to
detector array (21). with analytes on the waveguide surface. A pho- configuration and dispersive elements, simul-
Furthermore, the resolution will be affected todetector array then detects the light upon taneously decreasing the size and increas-
if there are no collimation components to its exit from an output grating. As such, the ing the sensitivity of the photodetector arrays
image the input slit onto the detector. A fea- system can be used to measure incident light provides another route toward smaller, higher-
sible approach is to use a concave grating spectra or the absorption spectra of the analytes performance microspectrometers. For exam-
(6, 22, 23); light is then dispersed and focused on top of the waveguide through evanescent ple, superconducting nanowires have recently
from different angles to different positions on coupling. emerged as one of the most promising alter-
the detector array without the need for com- These waveguide-based spectrometers have natives to semiconductor photodetectors for
plex collimation optics and multiple reflective used various dispersion schemes (Fig. 2, E to spectrometer applications, as they show ultra-
components (Fig. 2B) (23). Commercial, man- I) such as planar photonic crystals (18, 26), high sensitivities (single-photon detectability)
ufacturable, visible-range microspectrometers holographic elements (19, 27), planar echelle with low jitter and dark counts (11, 12). Indeed,
based on this design have achieved a resolu- gratings (12, 28), transmission gratings (29), recent single-photon microspectrometer dem-
tion of ~10 nm with a footprint of 1 to 2 cm self-focusing transmission gratings (16), chirped onstrations using superconducting nanowires
(23). In addition to these concave gratings, gratings (30), and arrayed waveguide gratings are capable of carrying out spectral analysis of
meta-lenses (24) and grating-Fresnel lenses (AWGs) (17), as well as metasurfaces (31). As ultrafaint light, which marks a step toward
(25) have been demonstrated as diffractive with the out-of-plane devices, the resolution of their use in astronomical spectroscopy and
optical elements. The grating-Fresnel lens— waveguide-based spectrometers is inherently quantum computing (11, 12).
an integrated combination of a diffraction tied to the optical path length afforded by the
grating and Fresnel lens—has also been de- system’s footprint, and as such, miniaturiza- Narrowband filters
veloped within a smartphone-attached spec- tion necessitates a reduction in performance. Narrowband filters selectively transmit light
trometer system (Fig. 2C) (9). Furthermore, with respect to manufacturabil- with specific wavelengths, allowing for their
Waveguides have been proposed as an al- ity, fabrication tolerances (for example, in rela- use in spectrally selective detection. In these
ternative to free-space optics, allowing more tion to sidewall roughness–induced losses) systems, light dispersion can be achieved either
compact light confinement, to further reduce and waveguide mode coupling at low chan- with a single filter, the transmissive properties
footprint without considerable compromise on nel spacings present a challenge for extreme of which can be varied over time, or by pass-
performance (14). For this approach, input and miniaturization below the millimeter scale ing light through an array of multiple unique
output gratings are etched on the two sides (32, 33). However, aside from greater light narrowband filters each mounted onto their

Yang et al., Science 371, eabe0722 (2021) 29 January 2021 2 of 11

RES EARCH | R E V I E W

A Grating Detector array B Input slit C Input slit

Detector Reflective concave

array blazed grating
Grating-Fresnel
Detector lens
Mirror array

D Planar waveguide
E F Planar holograms

Outputs
Input
Detector array
Grating / metasurface Input

Substrate Photonic crystal Output channels

G Input Planar H Transmission I Waveguides ΔL

waveguide grating waveguide grating

Detector Detector Outputs

array array Input

Free propagation region

Downloaded from https://www.science.org on December 29, 2023

Fig. 2. Spatially dispersive microspectrometers. (A and B) Miniaturized different dispersion elements: (D) a buried grating on a waveguide sensor (14),
spectrometer systems based on out-of-plane spatial dispersion using planar (21) (E) a photonic crystal–based grating (18), (F) a holographic element (19),
and concave (23) gratings, respectively. (C) Schematic of a grating-Fresnel (G) a planar echelle grating (28), (H) a transmission waveguide grating (16), and (I)
spectrometer (9). (D to I) Waveguide-based spectrometers based around an arrayed waveguide grating (17). [(F) adapted with permission from (19).]

own detector. Although they are still limited A typical structure for a tunable Fabry-Pérot electro-optically tunable LiNbO3 (43, 44) or
by the detector and filter size, narrowband filter–based microspectrometer (Fig. 3A) fea- liquid crystals (45) as the cavity medium; the
filter-based spectrometers offer a key advan- tures a resonant optical cavity consisting of angle of incidence Dq has been varied via ro-
tage with respect to miniaturization over those two parallel mirrors separated by a variable tating the filter (46).
based on dispersive systems. Aside from the distance d. The transmission function T of There are certain key factors to consider in
benefit of their planarity, no separation (that the Fabry-Pérot filter is given by the Airy engineering these Fabry-Pérot spectrometers
is, path length) is required between the spectral function (41): for high performance. The FWHM of trans-
filtering element(s) and the detector(s), circum-
 mission peaks determines the spectral reso-
A 2
venting one of the fundamental limitations of 1  1r lution of the filter-based spectrometers, which
dispersive devices and affording the possibility T¼ h i ð1Þ in turn is equal
1 þ 4r 2 sin2 2p pffiffito
ffi the intrinsic finesse of the
of far more compact systems. ð1rÞ l ðnd cos qÞ  ϕ cavity Fint ¼ p r=ð1  rÞ (20). Thus, high res-
olution requires high reflectance, but for a
Tunable filter–based microspectrometers where A is the absorbance of mirrors and metallic mirror cavity, this results in lower
A range of tunable narrowband filters, such as cavity, r is the reflectance of the mirrors, n transmission and therefore a weaker signal-
acousto-optic tunable (AOTF) (34), liquid- is the refractive index of the cavity medium, q to-noise ratio (37). A partial solution here is
crystal tunable (LCTF) (35), and Fabry-Pérot is the incidence angle, and ϕ is the phase shift to use distributed Bragg reflectors as mirrors
(36–38) filters, as well as microring resonators at the reflectors, which normally is neglected. (47), which consist of alternating high– and
(39) have all been demonstrated in spectrom- Light can be resonated and enhanced in the low–refractive index dielectric quarter-wave
eters. Their spectral transmission can be rap- cavity when the optical distance between the layers with high reflectance and low absorp-
idly and dynamically controlled through the two mirrors (nd cos q) is an integral multiple tion at a specific spectral range. However, they
application of a voltage or acoustic signal that of its half wavelength l/2, which results in are costly and much more complex to manu-
temporally separates spectral components. maximum ideal transmission through the facture. Note that defects in the cavity system,
AOTFs use an acoustic field to generate a cavity (and reflector) to the detector (T = 1). including nonparallelism or mirror imperfec-
periodically fluctuating refractive index in Figure 3A shows T as a function of light wave- tions, also reduce the effective finesse (and
solid-state birefringent crystals, analogous length. Assuming that the absorbance and thus resolution):
to a tunable diffraction grating. However, to reflectance are fixed for a given device, trans-
date, size constraints on the birefringent crys- mission spectra can be tuned during operation 1 1 1
¼ þ ð2Þ
tals have presented a major obstacle to AOTF by varying the optical path length, which can F 2eff F 2int F 2D
spectrometer miniaturization. LCTFs suffer be achieved by changing n, d, or q. Tuning
from a similar constraint, containing a stack of the separation of the mirrors (Dd) is the most where FD represents the defect finesse (20).
polarizers and liquid crystal cells that are chal- common strategy and is achieved through the
lenging to miniaturize. On the other hand, tun- use of an electrostatic or piezoelectric actuator Filter arrays and linear variable filters
able Fabry-Pérot filters can be fabricated by (Fig. 3, B and C), where current MEMS tech- For the tunable narrowband filter spectrom-
well-established MEMS-compatible processes nology straightforwardly affords a wide tun- eters discussed above, the spectra are analyzed
(40), which make them highly suitable for mi- able range (36, 37, 41, 42). Tunability of the in a time sequence, sacrificing time response.
crospectrometer mass production (4, 37). refractive index Dn can be realized by choosing This also presents an obstacle for high-speed

Yang et al., Science 371, eabe0722 (2021) 29 January 2021 3 of 11

RES EARCH | R E V I E W

A I0 Reflectors B Flexible suspension Input Movable use of an interferometer to modulate the light
structure Mirrors Si frame incident on a single detector over time. The
Medium
d (n) “interferograms” collected at the detector
V Spacers
Si substrates
(functions of received signal intensity over
It time, or a time-variant property of the system
Control electrodes Output
100%
such as optical path length) are then converted
Membrane to a wavelength-dependent spectrum via FT.
FWHM C
T

V FT systems have two main inherent benefits

Si substrates
0% over those based on diffractive optics. First,
λ
collecting spectral information at one detec-
tor simultaneously results in the multiplex (or
F-P
D etalons E Output to detector array Fellgett’s) advantage. Second, avoiding spatial
Thin dispersion results in a higher optical through-
Filter film Resonator put, or étendue, known as Jacquinot’s advan-
array Photonic tage. Both of these factors tend toward affording
crystal a higher signal-to-noise ratio. Moreover, using
Input Waveguide Through
Detector array Metasurface one detector offers a smaller and more cost-
effective alternative to array-based detectors.
Miniaturized FT spectrometers can be cate-
F Input G gorized through the mechanism by which the
Tapered waveguide Cutoff λ optical path lengths within the interferometer
Mirrors
are changed over time; the broadest differ-

Downloaded from https://www.science.org on December 29, 2023

Tapered cavity layer entiation is between those with and without
Input moving parts. Belonging to the former group,
Output
the earliest chip-based FT spectrometers, ap-
Output
Photodetector array pearing first in the late 1990s, were based
around Michelson interferometers, with MEMS
Fig. 3. Narrowband filter spectrometers. (A) Schematic configuration (top) and the corresponding used to manipulate mirrors with either elec-
transmittance function (bottom) of a typical Fabry-Pérot filter. (B and C) Typical configurations of trostatic (62–64), electromagnetic (65), or elec-
tunable Fabry-Pérot filter–based microspectrometers showing a bulk wafer bonding structure (42) and a trothermal (66) (often comb-drive) actuators
membrane-on-wafer structure (4), respectively. (D) Schematic of a filter array microspectrometer scheme (Fig. 4A). Aside from difficulties in integration
based around Fabry-Pérot etalons (49), etched thin films (51), photonic crystals (52), and metasurface with planar light sources, a drawback of these
(54). (E) A waveguide ring resonator–based filter strategy. (F and G) Schematic of linear variable filter–based devices is that the spectral resolution is lim-
spectrometer designs and their implementation, for transverse incidence (60) and waveguided operation ited by the maximum optical path length dif-
(61), respectively. [Metasurface image at lower right of (D) adapted with permission from (54).] ference (OPD) allowed by the actuator travel
range (67), which in turn is constrained by, for
instance, the pull-in effect (whereby, below a
spectroscopy applications. Narrowband filter can be fabricated in only nine deposition pro- threshold separation, the mirrors will uncon-
arrays and linear variable filters offer an ad- cesses (51). trollably and rapidly attract each other) (2, 6).
vantage in allowing simultaneous measurement Another strategy to increase the ease of Very recently, it has been demonstrated that
of multiple spectral components in parallel fabricating a wide range of transmission wave- the evaporation of a droplet atop the end of
(although this in turn necessitates multiple lengths is to use a linear variable filter, which is an optical fiber can also function as a sys-
detectors). typically either a wedged (55, 56) or composition- tem analogous to a scanning FT spectrome-
Fixed filter arrays have been exploited in graded filter (57), where the transmission or ter, to obtain the absorption spectra of liquid
many microspectrometers, where each filter reflection spectrum varies continuously along analytes (68).
is responsible for transmitting a specific wave- one axis of the filter. One approach here is to As of the late 2000s, planar on-chip FT
length onto the photodetector underneath. simply scan a spectrum by sliding the filter spectrometers—based on integrated wave-
Various filter schemes exist, differentiated by over a single detector (58). However, this is guides and without movable mirrors—have
their working principles, configurations, and relatively slow and necessitates extra moving emerged. Instead of Michelson architectures,
materials (48). Filter arrays based on Fabry- parts. Combining a linear variable filter with these systems are based around Mach-Zehnder
Pérot etalons (49, 50), thin films (51), planar a detector array for parallel measurement interferometers (MZIs), whereby light is split
photonic crystals (52), photonic bandgap fi- offers an ideal solution (59). Microspectrom- into at least two unidirectional pathways
bers (53), metasurfaces (54), and waveguide eters based on this configuration have already toward a single detector where they are re-
ring resonators (39) have been demonstrated been demonstrated (Fig. 3F) (60). Tapered combined. Here, the OPD induces a phase dif-
for the development of microspectrometers Bragg waveguides exhibit similar linear varia- ference between light in different channels.
(Fig. 3, D and E). Clearly, the number of chan- ble filtering capability due to the variable cut- Some of these systems are based around arrays
nels directly constrains the spectral resolution off propagation wavelength (Fig. 3G) (61). Both of multiple MZIs, forming spatial heterodyne
[in the case of the 16 filters in (49), for exam- linear variable configurations show high spec- spectrometers (69–71). For instance, an array
ple, to ~25 nm], so processes that necessitate tral resolution (~1 nm). of spirally coiled waveguides was fabricated
the individual placement or deposition of each (Fig. 4B), with the length of the waveguides
filter are ill-suited. To address this, a combi- Fourier transform microspectrometers varying linearly by DL, inducing a delay be-
natorial deposition technique can be used to Typically used for absorption or emission spec- tween the paths (70). However, such a sys-
fabricate a large number of filters in fewer steps; troscopy in the infrared range, Fourier trans- tem is limited by the maximum OPD as well as
for example, an 8 × 16 array (128 channels) form (FT) spectrometers center around the the number of MZIs, both of which constrain

Yang et al., Science 371, eabe0722 (2021) 29 January 2021 4 of 11

RES EARCH | R E V I E W

A Stationary C such that, even with a fixed array of nanosam-

mirror Photodetector plers, the whole interferogram can be sampled,
Light Heater 2 in this case achieving a spectral range of 500 nm
input Heater 1
(79). The resolving power, R = l/Dl, in these
MZI Output devices is given by 2nL/l, where n is the re-
Beam
splitter Scanning
mirror fractive index of the waveguide, L is the length
Detector
VA VB Tunable ring of waveguide being probed, and l is the wave-
Electrostatic actuator resonator (resolution enhancement) length. As such, extremely high resolutions
(tens of picometers) are possible over only
Evanescent field samplers
centimeter length scales, although, conversely,
B D
Input Waveguide such devices may not be suitable for extreme
miniaturization. These systems also currently
Input light Interferogram
rely on an external camera to image the inter-
Substrate
3 μm Undersampled ferogram scattering from the samplers; the
development of specialized nanoscale photo-
ΔL1
detectors, to be placed on top of the wave-
V=0 guide, would allow for direct measurement
Electro-optic shift
and a simplified system.
1 mm

V≠0 Reconstructive spectrometers

-20 -15 -10 -5 0 5 10 15 20
Length (μm) Over the past decade, a new spectrometer
paradigm has emerged, which, as indicated by

Downloaded from https://www.science.org on December 29, 2023

Fig. 4. Fourier transform microspectrometer platforms. (A) Schematic of a miniaturized MEMS Fourier
the name, relies on computational techniques
transform spectrometer based on a Michelson interferometer. The path lengths of the interferometer are
to approximate or “reconstruct” an incident
varied over time by the electrostatic actuator to collect an interferogram, which is subsequently converted to
light spectra from precalibrated information
spectral data by Fourier transform (62). (B and C) MZI-based, on-chip Fourier transform spectrometers.
encoded within a set of detectors. More spe-
(B) Optical image of a spatial heterodyne spectrometer device based on an array of multiple MZIs with
cifically, the “reconstruction” here refers to the
different OPDs, each varying by DLi. (C) Schematic of a microspectrometer based on a single MZI,
solution of a linear equation system. Thus far,
using a microring resonator for resolution enhancement and a heater to facilitate path length modulation by
two strategies have generally been seen for
thermo-optical effect. [(B) and (C) adapted with permission from (70, 74), respectively.] (D) Schematic
encoding spectral information within a set of
(top) of a standing-wave integrated Fourier transform spectrometer system in an electro-optically tunable
detectors: complex spectral-to-spatial mapping
medium; diagrams (below) illustrate the use of the electro-optic effect in LiNbO3 to shift the interferogram
and spectral response engineering.
along the waveguide over time through application of a bias voltage V. [Adapted with permission from (79).]
Complex spectral-to-spatial mapping
the extent to which the spectrometer footprint wave integrated Fourier transform spectrom- In a conventional grating-based spectrometer,
can be reduced while maintaining satisfactory etry (SWIFTS). In recent works on SWIFTS- a point (i.e., a wavelength) in the spectral do-
performance. based microspectrometers, a standing wave main is mapped to a point (i.e., a detector) in
Building on this, alternative designs use is set up within a single-mode, closed-loop the spatial domain (80). The readout of the
mechanisms that can continuously tune the waveguide through the interference of two detectors directly constitutes the spectrum.
characteristics of a single MZI (and thus re- counterpropagating signals (alternatively, in However, as mentioned previously, the spec-
quire only a single photodetector) rather than a Lippmann configuration, a mirror can be tral resolution scales with the distance from the
relying on multiple MZIs with fixed lengths. placed at the end of a waveguide to set up a grating to the detectors (i.e., the path length);
This has been achieved through electro-optical standing wave by reflecting the signal back one-to-one spectral-to-spatial mapping is thus
modulation in, for example, LiNbO3 wave- upon itself) (77, 78). Whereas temporal inter- highly limited when looking to increase spec-
guides (67, 72) as well as through exploiting ferograms are produced and collected at a tral resolution with a decreasing footprint (81).
thermo-optical effects, where microheaters are single detector in the previously mentioned FT Complex spectral-to-spatial mapping is an al-
embedded adjacent to integrated optical path- spectrometers, here a spatial interferogram is ternative approach that distinguishes the
ways (73, 74) (Fig. 4C). An alternative approach produced. Metallic nanoribbons are deposited wavelengths by creating a signature pattern
is a digital FT spectrometer, where a photonic with a regular spacing on top of the waveguide (either one- or two-dimensional) in the spatial
circuit is used, featuring a number of optical to sample the evanescent field and map the domain for each of the wavelengths (Fig. 5A).
switches that divert the signal along paths of relative intensity of the standing wave along For example, when monochromatic light
different lengths (75). Here, the resolution de- the loop (Fig. 4D). A proof-of-concept demon- passes through a dispersive element such as
pends on the number of paths, or spectral stration for this design acknowledged that the a long multimode fiber (MMF), it will create
channel count, which scales exponentially with spectral range in such a system is constrained a wavelength-dependent signature pattern at
the number of optical switches. In many of (in this case, to 96 nm centered at 1500 nm), as the output of the fiber due to the interference
these methods, recent advances have allowed the samplers cannot be fabricated at a pitch to between the guided modes in the fiber (80).
computational techniques such as compres- avoid undersampling of the interferogram (77). Thus, when passing an arbitrary polychromatic
sive sensing (71), machine learning (76), and However, more recently, a SWIFTS system light through the MMF, the output will be the
forward-backward linear prediction (67) to has been developed that circumvents this issue overlay of scaled signature patterns created by
enhance the spectral resolution of these de- using a dual spatial and temporal sampling each of the individual wavelengths. The target
vices and correct for temperature change– or scheme. Use of the electro-optic effect in a hy- spectrum to be reconstructed is essentially the
fabrication-based errors. brid LiNbO3-SiN waveguide platform allows assembly of scaling weights corresponding to
A related variant of these miniaturized the spatial interferogram to be shifted along these patterns (81). The signature patterns can
interferometer–based FT systems is stationary the waveguide by applying a voltage (Fig. 4D) also be generated by feeding the light into a

Yang et al., Science 371, eabe0722 (2021) 29 January 2021 5 of 11

RES EARCH | R E V I E W

A B Detection C
Input: s( ) channels Detectors

Target pattern
Dispersive media
λ2 Disordered
structure Spiral waveguide
Detector array p ( ,x) . s( ) d = I(x)
λ1
D Hole array E Polychromator

Reconstructed
s( )

F I1 I1 λ2 G Quantum H Photonic
D1( ) . s( ) d = I1 dot filters crystal slabs
I2 I2 λ1
s( )

λ2
or D2( ) . s( ) d = I2
...

...

λ1
Ii Ii

...
λ2
D i( ) D i( )
Di( ) . s( ) d = Ii I J
Filter array Detector array λ1 Graded nanowire
Structurally
colored
... Reconstructed
Ii nanowires

Downloaded from https://www.science.org on December 29, 2023

n+ Si
s( ) n- Si
D1 ( ) D 2( ) Di ( ) p+ Si I1 I2 ... Ii-1 Ii
Spectral properties of elements Au
λ1 λ2

Fig. 5. Computational microspectrometers. (A) Operational principles behind engineered spectral response filtering or detection. (G to J) Examples of different
spectrometers based on spectral-to-spatial mapping. (B to E) Examples of reconstructive systems using spectrally engineered filtering, with arrays based
optical systems that had been adopted for complex spectral-to-spatial mapping: on colloidal quantum dot mixtures (G) and photonic crystal slabs (H), and
(B) a disordered photonic chip, (C) a spiral waveguide, (D) a dispersive hole spectrally engineered detection, based on arrays of structurally colored
array, or (E) a polychromator. [(B) to (E) adapted with permission from (81, 83–85), nanowires (I) and a single compositionally engineered nanowire (J). [(G) to (J)
respectively.] (F) Principles behind reconstructive spectrometers based on adapted with permission from (86), (88), (95), and (13), respectively.]

miniaturized optical element such as a dis- as possible. It has been recognized that the response for each of the detectors. This can be
ordered photonic chip (81, 82), a spiral wave- diversity is proportional to the spread of the achieved either by engineering the detectors
guide (83), a dispersive hole array (84), or a optical path length (i.e., the difference be- themselves or the optical elements integrated
polychromator (85) (Fig. 5, B to E). For a tween the shortest and longest optical paths on top of the detectors (Fig. 5F). Assuming the
sought spectrum s(l) with normalized sig- of the propagation modes within the chosen spectral response of the ith detector (or the
nature patterns p(l, x) and signals measured optical elements) (81). A resolution of 0.01 nm filter-detector pair) to be Di(l), its measured
at the detector array I(x), the mapping process has been demonstrated with a multimode signal, Ii, can be described as
can be mathematically described as spiral waveguide that uses evanescent cou-
pling to considerably enhance the optical path l2
l2 Ii ¼ ∫l1 Di ðlÞ  sðlÞdl ð5Þ
IðxÞ ¼ ∫l1 pðl; xÞ  sðlÞdl
length spread (83). Such a high resolution can
ð3Þ greatly broaden the applications of miniatur-
ized spectrometers. In addition, these sys- The signals of the detector array can then be
where x is a vector denoting a position on tems are robust with respect to fabrication summarized in a discrete format as
the detector array. Equation 3 can be dis- imperfections, which can be compensated for
I¼DS ð6Þ
cretized as through calibration (83). However, they suf-
I¼PS ð4Þ fer from temperature variations, which can where D is a matrix, the rows of which cor-
change the signature pattern for a specific respond to the detectors’ spectral response
where I, S, and P are the measured intensity, wavelength. The higher the spectral resolu- functions.
spectrum, and mapping matrices, respectively. tion, the more the spectrometer suffers from By solving Eq. 6, the unknown target spec-
If the signature patterns of two distinct thermal instability. Thus, additional measures trum can be reconstructed. The entire process
wavelengths are identical, it is impossible to such as adding a temperature controller or is illustrated by Fig. 5F. Almost any optical
tell which wavelength leads to the measured temperature-dependent calibration should be element that can generate diverse spectral re-
pattern. In this way, the dissimilarity between adopted (83). Furthermore, the computational sponse functions can be adopted for spectrom-
the signature patterns determines the resolv- cost scales up with the spectral resolution etry systems based on this principle, such as
ing power of the reconstructive spectrometers. for a fixed spectral range, as more variables quantum dots (86, 87) and photonic crystal
Mathematically, the signature patterns are need to be solved from an increased number slabs (88) (Fig. 5, G and H). Other designs
the columns of the matrix P; the higher the of equations. including liquid crystals (89), thin films (90),
similarity between the signature patterns, the etalon arrays (91), nanostructured photodiodes
larger the condition number of P, leading to a Spectral response engineering (92, 93), and metasurfaces (94) have also been
poorer numerical solution of Eq. 4. As a result, The second approach to realizing a reconstruc- used as the basis for such strategies. However,
the signature patterns should be as diverse tive spectrometer is to tailor a distinct spectral all these designs require separate fabrication

Yang et al., Science 371, eabe0722 (2021) 29 January 2021 6 of 11

RES EARCH | R E V I E W

of filter and detector arrays. This increases the between the functions is minimized (88). consideration. In this respect, there are two
complexity of manufacturing and limits min- Dictionary/machine-learning techniques (104) main impinging factors: changes in tempera-
iaturization. Recently, spectral responsivity– are also promising alternatives to incorporate ture and in air composition (for instance, the
engineered nanostructure has been demon- prior knowledge into the spectral reconstruc- impact of moisture). Thermal effects are of
strated to integrate both of these functions tion process. particular importance where the system is high-
(13, 95–99), as shown in Fig. 5, I and J. For ly sensitive to changes in the refractive index
example, a computational spectrometer based Summary and discussion of the active media, for example, in integrated
on a composition-gradient alloyed semicon- We have summarized the technological evolu- systems such as AWGs, MZIs, and SWIFTSs.
ductor nanowire was proposed, which can be tion of miniaturized spectrometers, detail- Likewise, challenges can arise from the thermal
divided into a number of sections (detectors) ing their working principles and merits, under expansion of gratings or MEMS components,
along the axial direction (13). The response four broad classifications. Despite sharing sim- as well as through temperature dependencies
functions of these detectors vary as a result of ilar overarching aims, works from these four in the spectral response function of the detec-
the gradual and subtle variation in the alloy subfields encompass a hugely diverse range tors in computational spectrometers. In many
compositions. Although resolution is still mod- of device designs and operational strategies. cases, temperature-sensitive calibration can
est (~5 to 10 nm), such a nanowire spectrometer Drawing together the field as a whole, Fig. 6A be applied to effectively eliminate (108) such
integrates the functions of both wavelength compares the resolution, operational wave- effects, and hermetic sealing (3) or passivating
selectivity and photodetection into an indi- length range, and footprint across the four coatings (13) can also be applied to stabilize
vidual nanostructure, pushing the footprint subcategorizations of device strategies defined device performance.
toward tens of micrometers, two orders of above. Clearly, there is a wide variance in per- Finally, the relative maturity of these sub-
magnitude below that of any other computa- formance, footprint, and operational range— fields must be taken into account when con-
tional spectrometer system. even within each subfield—and no platform sidering their prospects. To this end, Fig. 6B
However, as the size of the spectrometer yet combines high resolution (<1 nm), wide displays a timeline illustrating the emergence

Downloaded from https://www.science.org on December 29, 2023

decreases, the number of detectors (i.e., the spectral range (>300 nm), and ultracompact of key design innovations. As may be expected,
number of equations) that can be accommo- physical dimensions (<100 mm). Behind these these milestones have followed wider techno-
dated also decreases because of the physical metrics, it is important to also consider the logical trends; early microspectrometers mainly
constraints, affecting the ultimate spectral suitability and ease of integration of each de- took advantage of breakthroughs in micro-
resolution that can be achieved. In addition, vice for different application spaces. For in- fabrication between the 1980s and the early
the minimized footprint of the spectrometer stance, although the MEMS-based system in 2000s; advances in lithographic and etch pro-
reduces the light-matter interaction, which (105) shows competitive resolution perform- cesses, as well as the continued development of
compromises sensitivity. In a follow-up work, ance, wide spectral range, and a small foot- MEMS technology and waveguide-based chips,
a nanowire spectrometer was developed that print, its complex, 400-mm-tall structure would afforded the production of complex miniatur-
can operate in a waveguide mode, providing prove highly challenging for integration into ized dispersive, Fourier transform, or filter-
an improved signal-to-noise ratio for such an spectrometer arrays such as those required based systems.
ultracompact device (96). in the snapshot hyperspectral imaging cam- However, in the past decade, sharp increases
eras discussed below. in computational power and reductions in pro-
Reconstruction techniques Additionally, the sensitivity of each device cessor price and size have seen attention shift
The inversion problems presented above are design is an increasingly important factor as toward spectral reconstruction schemes based
typically ill-posed. Different strategies must the detector, and thus the photon collection on relatively simple, and often disordered, de-
be adapted to alleviate this, depending on area, is further reduced in size. This is espe- vice frameworks. We believe these systems rep-
whether the problem is overdetermined or cially noteworthy given that many portable resent the most promising paradigm, as their
underdetermined—that is, whether the num- applications will rely on the collection of am- performance can be improved not only by
ber of detectors (and thus, spectral response bient light. Here, there are many common fac- augmenting their hardware but, sometimes
functions) is higher or lower than the number tors among the four categories. For instance, more straightforwardly, by optimizing the soft-
of data points in the reconstructed spectrum, the nonintegrated systems discussed above can ware that powers them. It seems likely that this
respectively. In the overdetermined case, noise benefit from the introduction of a lens to focus trend will continue through further develop-
in the measurements will be readily amplified light onto the detectors, filter array, or disper- ment and optimization of machine learning–
into reconstruction errors (100). To mitigate sive medium (86, 106); enhancing the signal-to- based techniques, where the computational
this, truncated singular value decomposition noise ratio of the detectors is also one clear power of the accompanying processing sys-
can be adopted to remove some part of the avenue for improvement (5). However, there tems can shoulder much of the burden for en-
measured information that is most easily af- are also a variety of distinct considerations hancing spectral resolution. As these spectral
fected by the noise (101). When the problem dependent on the device design. In AWG sys- reconstruction algorithms mature, they will
is underdetermined, additional information tems, for instance, the overall sensitivity is increasingly be able to compensate for the
such as smoothness should be incorporated heavily influenced by the efficiency of coupling compromises in detector performance neces-
to find the most “plausible” solution based light into the waveguide, as well as the disper- sitated by further miniaturization, allowing for
on prior knowledge of the type of spectrum sion and propagation losses (15). As discussed ultracompact yet high-performance systems.
being measured (102). For example, the origi- previously, although FT systems benefit from Clear hurdles still remain in this respect. For
nal spectrum can be approximated as the com- a multiplex advantage in the use of only one example, deep learning algorithms typically
bination of a set of smooth basis functions such detector, transmittance from, for instance, a require very large, labeled datasets for proper
as Gaussian curves. Compressive sensing has Fabry-Pérot cavity can be as low as 15% of the training of the neural networks used, in order
also been adopted to explore the sparseness incident light (107). to establish an accurate relationship between
of the sought spectrum (103). According to Given the need for these devices to function the measurements and the spectrum to be
compressive sensing theory, randomly struc- outside controlled laboratory settings, their reconstructed. In cases where it is challeng-
tured response functions are preferred to im- robustness and stability with respect to their ing to produce sufficient high-quality training
prove the spectral resolution as the correlation external environment is another operational data, recent developments in transfer learning

Yang et al., Science 371, eabe0722 (2021) 29 January 2021 7 of 11

RES EARCH | R E V I E W

A 100 F-P tunable* (42)

Waveguide grating (29) MEMS (64)

F-P array (49)

F-P tunable* (38)
Single nanowire (13)
μ-grating* (23)
10
Resolution (nm)

MEMS (105)
μ-grating (21) Single photon grating (12)
Electro-optic SWIFTS (79)
Quantum dots (86)
Linear variable (60)
Filter array (90) Photonic crystal slab (88)
1 Photonic crystal (26) Ring resonator (39)
Disordered photonic chip (81)
Thermo-optic MZI (74)
Meta-lens (24)
Digital MZI (75)
Holographic (27)
0.1 AWG (17)

400 600 800 1000 1200 1400 1600 1800 2000 3 4 5

Spectral range (nm) Spectral range (μm)

Downloaded from https://www.science.org on December 29, 2023

Out of plane diffraction In plane diffraction Narrowband filter < 100 μm 0.1 - 1 mm
Footprint:
Fourier transform Reconstructive * Commercialized 1 - 10 mm 1 - 2.5 cm

B
Detector array
ΔL1
ΔL2
Input

ΔL3
...

ΔLi Metamaterial
Disordered array snapshot
μ-grating (21) MEMS-FT (62) MZI array (69) photonic (81) μ-grating* (22) imaging (117)
Ring resonator filter (39) μ-grating* Digital
(23) MZI (75)
Δn liquid crystal Linear ΔL tunable Photonic Tunable Single photon
tunable filter (45) variable (56) filter* (112) AWG (17) crystal (26) MZI (72) AWG (11)

1980s 1990s 2000 2005 2010 2015 2020

Linear ΔL tunable Waveguide Δ tunable SWIFTS (77) Grating-Fresnel (25)Single photon Electro-optic
variable (58) filter (36) grating (29) filter (46) grating (12) SWIFTS (79)
Broadband Holographic MEMS-FT (111)
filter array (100) (19) *

Quantum dot Single nanowire

μ-grating on Filter array ΔL tunable filter* broadband filter broadband
waveguide (14) (50) (42) array on CCD (86) detector array (13)

Fig. 6. The field of miniaturized spectroscopic devices. (A) Plot comparing numbers are indicated within parentheses. Note that the spectral ranges for (17),
the resolution, operational spectral range, and footprint for selected device (26), (75), and (81) (which are obscured by the resolution data points) are
demonstrations in the literature and those that are commercially available 10 nm, 10 nm, 20 nm, and 25 nm, respectively. (B) Timeline illustrating
(indicated by asterisks), as categorized into their respective subfields the emergence of different technological platforms for microspectrometer
(see color key). Footprint encompasses those elements of the device that systems from the 1980s to the present day, sorted by subfield as displayed in
are active in resolving and detecting light, and does not include accessory the color key in (A). [Images either redrawn or adapted with permission
components such as the readout electronics or packaging. Reference from the references indicated.]

(109, 110) may provide a solution. Here, knowl- must progress in tandem with design of these mental sensitivity must be carefully controlled.
edge learned from a different but related task algorithms, taking into consideration the fac- As made evident by the equations above, more
is used to improve the performance of the tors that are most detrimental to their per- accurate reconstructions should be possible by
deep learning algorithm for a target task where formance. In reconstructive spectrometers, a increasing the number of detectors and the
sufficient training data are not available. key issue is noise—that is, inconsistencies be- diversity of the spectral response functions,
As well as producing more powerful soft- tween like measurements of the same spectrum. both of which present their own engineering
ware, further optimization of device hardware As discussed previously, thermal and environ- challenges for consideration in the device

Yang et al., Science 371, eabe0722 (2021) 29 January 2021 8 of 11

RES EARCH | R E V I E W

architecture. Furthermore, for commercial Perhaps most exciting are the possibilities 13. Z. Yang et al., Single-nanowire spectrometers. Science 365,
viability, miniaturization of the charge-coupled for hyperspectral imaging applications in- 1017–1020 (2019). doi: 10.1126/science.aax8814;
pmid: 31488686
device or complementary metal-oxide semicon- volving the simultaneous capture of spec- 14. D. S. Goldman, P. L. White, N. C. Anheier, Miniaturized
ductor sensors that accompany the filter arrays, tral and spatial information in a “data cube” spectrometer employing planar waveguides and grating
as well as associated readout and processing with dimensions (x, y, l). Here a long-term couplers for chemical analysis. Appl. Opt. 29, 4583–4589
(1990). doi: 10.1364/AO.29.004583; pmid: 20577437
electronics, must also be factored in. goal is a miniaturized, portable “snapshot” 15. A. Z. Subramanian et al., Silicon and silicon nitride photonic
spectral imager with high spectral and spatial circuits for spectroscopic sensing on-a-chip. Photon. Res. 3,
Application outlook resolution—that is, a camera whereby each B47 (2015). doi: 10.1364/PRJ.3.000B47
16. D. Sander, J. Muller, Selffocussing phase transmission grating
As also seen in Fig. 6, some platforms have pixel holds its own high-performance spec- for an integrated optical microspectrometer. Sens. Actuators
already reached full commercial development, trometer. Such a system would be a marked A 88, 1–9 (2001). doi: 10.1016/S0924-4247(00)00499-4
showcasing market demand for microspec- advance on many current strategies in devel- 17. P. Cheben et al., A high-resolution silicon-on-insulator
arrayed waveguide grating microspectrometer with
trometer systems. Over the past two decades, opment, which involve scanning systems with
sub-micrometer aperture waveguides. Opt. Express 15,
miniaturized, MEMS-based tunable filters movable parts, or digital mirror device–based 2299–2306 (2007). doi: 10.1364/OE.15.002299;
and Fourier transform spectrometers as well spatial light modulators (116). Given their sim- pmid: 19532464
as centimeter-scale, packaged, grating-based ple (and usually planar) device frameworks, as 18. B. Gao, Z. Shi, R. W. Boyd, Design of flat-band superprism
structures for on-chip spectroscopy. Opt. Express 23,
systems have become commercially available the active sensing area of computational micro- 6491–6496 (2015). doi: 10.1364/OE.23.006491;
(3, 22, 23, 42, 111, 112). These systems can all be spectrometers shrinks toward the ~10 mm scale, pmid: 25836867
produced with well-established semiconduc- miniaturized snapshot imagers based on such 19. G. Calafiore et al., Holographic planar lightwave circuit for
on-chip spectroscopy. Light Sci. Appl. 3, e203 (2014).
tor device fabrication infrastructure; emerging concepts become increasingly feasible. Indeed, doi: 10.1038/lsa.2014.84
systems that also fit into these frameworks successful prototypes based on arrays of “super- 20. R. F. Wolffenbuttel, State-of-the-art in integrated optical
(such as those based on photonic crystals) pixels,” each containing their own filter array, microspectrometers. IEEE Trans. Instrum. Meas. 53, 197–202
(2004). doi: 10.1109/TIM.2003.821490
have a clear advantage over those where novel have very recently been reported (117, 118). The

Downloaded from https://www.science.org on December 29, 2023

21. T. A. Kwa, R. F. Wolffenbuttel, Integrated grating/detector
manufacturing processes must be developed, addition of a diffuser here can convey a multi- array fabricated in silicon using micromachining techniques.
such as for nanomaterials-based devices. plexing advantage whereby each point in the Sens. Actuators A 31, 259–266 (1992). doi: 10.1016/
0924-4247(92)80114-I
More recently, the development of spectral object plane is mapped to many different 22. T. Yokino et al., Grating-based ultra-compact SWNIR spectral
sensing systems embedded within smart- points (rather than one point, as with a lens) sensor head developed through MOEMS technology.
phones has been evidenced, at CES (consumer at the superpixel array (118). Besides consumer Proc. SPIE 10931, 8 (2019). doi: 10.1117/12.2510472
23. Hamamatsu mini-spectrometer micro series C12666MA
electronics show) 2017 and in patent appli- technology, such devices could potentially re-
(2015); www.hamamatsu.com/jp/en/product/type/
cations by major smartphone manufacturers, volutionize existing hyperspectral imaging ap- C12666MA/index.html.
suggesting that an emergence into the public plications in agriculture and mineralogy. 24. A. Y. Zhu et al., Ultra-compact visible chiral spectrometer
domain may be imminent (113, 114). Indeed, with meta-lenses. APL Photon. 2, 036103 (2017).
doi: 10.1063/1.4974259
the prospect of a device with a footprint suit- 25. C. Yang, K. Shi, P. Edwards, Z. Liu, Demonstration of a PDMS
RE FERENCES AND NOTES
able for the smartphone paradigm is arguably based hybrid grating and Fresnel lens (G-Fresnel) device.
1. C. P. Bacon, Y. Mattley, R. DeFrece, Miniature spectroscopic Opt. Express 18, 23529–23534 (2010). doi: 10.1364/
the most important factor driving extreme instrumentation: Applications to biology and chemistry. OE.18.023529; pmid: 21164696
miniaturization of microspectrometer systems; Rev. Sci. Instrum. 75, 1–16 (2004). doi: 10.1063/1.1633025 26. B. Momeni, E. S. Hosseini, A. Adibi, Planar photonic crystal
a breakthrough demonstration here could prove 2. L. P. Schuler, J. S. Milne, J. M. Dell, L. Faraone, MEMS-based microspectrometers in silicon-nitride for the visible range.
microspectrometer technologies for NIR and MIR Opt. Express 17, 17060–17069 (2009). doi: 10.1364/
pivotal in terms of attracting further attention wavelengths. J. Phys. D 42, 133001 (2009). doi: 10.1088/ OE.17.017060; pmid: 19770924
and investment in this field. Although no re- 0022-3727/42/13/133001 27. A. Koshelev et al., Combination of a spectrometer-on-chip
constructive systems have reached commercial 3. J. Malinen et al., Advances in miniature spectrometer and and an array of Young’s interferometers for laser spectrum
sensor development. Proc. SPIE 9101, 91010C (2014). monitoring. Opt. Lett. 39, 5645–5648 (2014). doi: 10.1364/
maturity as yet, the readily available process- doi: 10.1117/12.2053567 OL.39.005645; pmid: 25360949
ing power, coupled with the need for systems 4. M. Ebermann et al., Tunable MEMS Fabry-Pérot filters for 28. J.-J. He et al., Monolithic integrated wavelength demultiplexer
with minimized footprint and weight, make infrared microspectrometers: A review. Proc. SPIE 9760, based on a waveguide Rowland circle grating in InGaAsP/InP.
97600H (2016). doi: 10.1117/12.2209288
the smartphone platform an obvious area for J. Lightwave Technol. 16, 631–638 (1998). doi: 10.1109/
5. R. A. Crocombe, Portable Spectroscopy. Appl. Spectrosc. 72, 50.664075
these microspectrometers to emerge. 1701–1751 (2018). doi: 10.1177/0003702818809719; 29. D. Sander, M.-O. Duecker, O. Blume, J. Mueller, An optical
Beyond smartphones, we believe that high- pmid: 30335465 microspectrometer in SiON-slab-waveguides. Proc. SPIE
performance spectrometry contained within 6. R. F. Wolffenbuttel, MEMS-based optical mini- and 2686, 100–107 (1996). doi: 10.1117/12.236128
microspectrometers for the visible and infrared spectral 30. S. Nezhadbadeh, A. Neumann, P. Zarkesh-Ha, S. R. J. Brueck,
ultraminiaturized packaged systems will find range. J. Micromech. Microeng. 15, S145–S152 (2005). Chirped-grating spectrometer-on-a-chip. Opt. Express 28,
applications in a vast range of fields, indus- doi: 10.1088/0960-1317/15/7/021 24501–24510 (2020). doi: 10.1364/OE.398072;
tries, and commercial technologies, including 7. M. Manley, Near-infrared spectroscopy and hyperspectral pmid: 32906990
imaging: Non-destructive analysis of biological materials. 31. M. Faraji-Dana et al., Compact folded metasurface
satellites and drones, wearables and implant- Chem. Soc. Rev. 43, 8200–8214 (2014). doi: 10.1039/ spectrometer. Nat. Commun. 9, 4196 (2018). doi: 10.1038/
able devices, chemical and food manufacturing, C4CS00062E; pmid: 25156745 s41467-018-06495-5; pmid: 30305616
and cellular imaging and lab-on-a-chip systems. 8. A. J. S. McGonigle et al., Smartphone Spectrometers. Sensors 32. S. Janz et al., Planar waveguide echelle gratings in silica-on-
18, 223 (2018). doi: 10.3390/s18010223; pmid: 29342899
The size of ultraminiaturized devices (<100 mm) 9. P. Edwards et al., Smartphone based optical spectrometer for
silicon. IEEE Photonics Technol. Lett. 16, 503–505 (2004).
doi: 10.1109/LPT.2003.823139
alone may make them suitable for wearable diffusive reflectance spectroscopic measurement of 33. J. Zou, T. Lang, Z. Le, J. J. He, Ultracompact silicon-on-
or flexible technology, given the comparative- hemoglobin. Sci. Rep. 7, 12224 (2017). doi: 10.1038/s41598- insulator-based reflective arrayed waveguide gratings for
017-12482-5; pmid: 28939898 spectroscopic applications. Appl. Opt. 55, 3531–3536 (2016).
ly large bending radii expected in operation. 10. E. Ryckeboer, R. Bockstaele, M. Vanslembrouck, R. Baets, doi: 10.1364/AO.55.003531; pmid: 27140366
Array-based devices that already feature flexi- Glucose sensing by waveguide-based absorption 34. H. Y. Zhang, X. L. Wang, J. Soos, J. Crisp, Design of a
ble, thin-film filters (86) could prove promising spectroscopy on a silicon chip. Biomed. Opt. Express 5, miniature solid-state NIR spectrometer. Proc. SPIE 2475,
1636–1648 (2014). doi: 10.1364/BOE.5.001636; 376–383 (1995). doi: 10.1117/12.211276
in this regard if the rigid detectors beneath
pmid: 24877021 35. N. Gat, Imaging spectroscopy using tunable filters: A review.
could be replaced with emerging, flexible coun- 11. V. Kovalyuk et al., On-chip single-photon spectrometer for Proc. SPIE 4056, 50–64 (2000). doi: 10.1117/12.381686
terparts (115). Furthermore, bandgap engineer- visible and infrared wavelength range. J. Phys. Conf. Ser. 1124, 36. S. R. Mallinson, J. H. Jerman, Miniature micromachined
ing of solution-processable materials (98, 99) 051045 (2018). doi: 10.1088/1742-6596/1124/5/051045 Fabry-Perot interferometers in silicon. Electron. Lett. 23, 1041
12. R. Cheng et al., Broadband on-chip single-photon (1987). doi: 10.1049/el:19870728
could offer a route toward printable, confor- spectrometer. Nat. Commun. 10, 4104 (2019). doi: 10.1038/ 37. J. P. Carmo et al., A review of visible-range Fabry-Perot
mable spectrometers. s41467-019-12149-x; pmid: 31506440 microspectrometers in silicon for the industry. Opt. Laser

Yang et al., Science 371, eabe0722 (2021) 29 January 2021 9 of 11

RES EARCH | R E V I E W

Technol. 44, 2312–2320 (2012). doi: 10.1016/ hollow Bragg waveguides. Opt. Express 17, 16632–16645 82. W. Hartmann et al., Waveguide‐Integrated Broadband
j.optlastec.2012.03.036 (2009). doi: 10.1364/OE.17.016632; pmid: 19770879 Spectrometer Based on Tailored Disorder. Adv. Opt. Mater. 8,
38. J. Antila et al., MEMS and piezo actuator-based Fabry-Perot 62. O. Manzardo, H. P. Herzig, C. R. Marxer, N. F. de Rooij, 1901602 (2020). doi: 10.1002/adom.201901602
interferometer technologies and applications at VTT. Proc. Miniaturized time-scanning Fourier transform spectrometer 83. B. Redding, S. Fatt Liew, Y. Bromberg, R. Sarma, H. Cao,
SPIE 7680, 76800U (2010). doi: 10.1117/12.850164 based on silicon technology. Opt. Lett. 24, 1705–1707 (1999). Evanescently coupled multimode spiral spectrometer. Optica
39. A. Nitkowski, L. Chen, M. Lipson, Cavity-enhanced on-chip doi: 10.1364/OL.24.001705; pmid: 18079909 3, 956 (2016). doi: 10.1364/OPTICA.3.000956
absorption spectroscopy using microring resonators. Opt. 63. M. Erfan et al., On-chip micro-electro-mechanical system 84. T. Yang et al., Miniature spectrometer based on diffraction in
Express 16, 11930–11936 (2008). doi: 10.1364/OE.16.011930; Fourier transform infrared (MEMS FT-IR) spectrometer-based a dispersive hole array. Opt. Lett. 40, 3217–3220 (2015).
pmid: 18679466 gas sensing. Appl. Spectrosc. 70, 897–904 (2016). doi: 10.1364/OL.40.003217; pmid: 26125406
40. P. M. Zavracky, E. Hennenberg, Miniature Fabry Perot doi: 10.1177/0003702816638295; pmid: 27044847 85. P. Wang, R. Menon, Computational spectrometer based on a
spectrometers using micromachining technology. 64. T. Sandner, A. Kenda, C. Drabe, H. Schenk, W. Scherf, broadband diffractive optic. Opt. Express 22, 14575–14587
Proc. WESCON 95, 325–332 (1995). doi: 10.1109/ Miniaturized FTIR-spectrometer based on optical MEMS (2014). doi: 10.1364/OE.22.014575; pmid: 24977553
WESCON.1995.485299 translatory actuator. Proc. SPIE 6466, 646602 (2007). 86. J. Bao, M. G. Bawendi, A colloidal quantum dot spectrometer.
41. J. H. Jerman, D. J. Clift, S. R. Mallinson, A miniature doi: 10.1117/12.697898 Nature 523, 67–70 (2015). doi: 10.1038/nature14576;
Fabry-Perot-interferometer with a corrugated silicon 65. U. Wallrabe, C. Solf, J. Mohr, J. G. Korvink, Miniaturized pmid: 26135449
diaphragm support. Sens. Actuators A 29, 151–158 (1991). Fourier Transform spectrometer for the near infrared 87. X. Zhu et al., Broadband perovskite quantum dot
doi: 10.1016/0924-4247(91)87117-L wavelength regime incorporating an electromagnetic linear spectrometer beyond human visual resolution. Light Sci.
42. N. Neumann, M. Ebermann, S. Kurth, K. Hiller, Tunable actuator. Sens. Actuators A 123–124, 459–467 (2005). Appl. 9, 73 (2020). doi: 10.1038/s41377-020-0301-4;
infrared detector with integrated micromachined Fabry-Perot doi: 10.1016/j.sna.2005.05.014 pmid: 32377335
filter. J. Micro-Nanolith. MEM 7, 021004 (2008). 66. W. Wang, S. R. Samuelson, J. Chen, H. Xie, Miniaturizing 88. Z. Wang et al., Single-shot on-chip spectral sensors based on
43. W. J. Gunning, P. Yeh, Multiple-cavity infrared electro-optic Fourier transform spectrometer with an electrothermal photonic crystal slabs. Nat. Commun. 10, 1020 (2019).
tunable filter. Proc. SPIE 0202, 21–25 (1980). doi: 10.1117/ micromirror. IEEE Photonics Technol. Lett. 27, 1418–1421 doi: 10.1038/s41467-019-08994-5; pmid: 30833569
12.958092 (2015). doi: 10.1109/LPT.2015.2423637 89. Y. August, A. Stern, Compressive sensing spectrometry
44. Y. Yao et al., Design of programmable multi-wavelength 67. K. Wang, J. Li, D. F. Lu, Z. M. Qi, Algorithmic enhancement of based on liquid crystal devices. Opt. Lett. 38, 4996–4999
tunable filter on lithium niobate. Results Phys. 15, 102741 spectral resolution of a LiNbO3 waveguide-based miniature (2013). doi: 10.1364/OL.38.004996; pmid: 24281493
(2019). doi: 10.1016/j.rinp.2019.102741 Fourier transform spectrometer. Appl. Spectrosc. 70, 90. J. Oliver, W. B. Lee, H. N. Lee, Filters with random
45. W. Gunning, J. Pasko, J. Tracy, A liquid crystal tunable 1685–1691 (2016). doi: 10.1177/0003702816644454; transmittance for improving resolution in filter-array-based
spectral filter: Visible and infrared operation. Proc. SPIE pmid: 27402687 spectrometers. Opt. Express 21, 3969–3989 (2013).

Downloaded from https://www.science.org on December 29, 2023

0268, 190–194 (1981). doi: 10.1117/12.959943 68. P. Malara et al., A self-operating broadband spectrometer on doi: 10.1364/OE.21.003969; pmid: 23481932
46. G. Lammel, S. Schweizer, P. Renaud, Microspectrometer a droplet. Nat. Commun. 11, 2263 (2020). doi: 10.1038/ 91. E. Huang, Q. Ma, Z. Liu, Etalon array reconstructive
based on a tunable optical filter of porous silicon. Sens. s41467-020-16206-8; pmid: 32385273 spectrometry. Sci. Rep. 7, 40693 (2017). doi: 10.1038/
Actuators A 92, 52–59 (2001). doi: 10.1016/ 69. M. Florjańczyk et al., Multiaperture planar waveguide srep40693; pmid: 28074883
S0924-4247(01)00539-8 spectrometer formed by arrayed Mach-Zehnder 92. A. Ahamad et al., Smart nanophotonics silicon spectrometer
47. J.-H. Liao, W. Wang, C.-J. Chen, C.-J. Yu, M.-C. Wu, Design interferometers. Opt. Express 15, 18176–18189 (2007). array for hyperspectral imaging. CLEO, STh3M.2 (2020).
and fabrication of large-area tunable MOEMS-based doi: 10.1364/OE.15.018176; pmid: 19551116 doi: 10.1364/CLEO_SI.2020.STh3M.2
shortwave infrared Fabry-Pérot filters. J. Vac. Sci. Technol. B 70. A. V. Velasco et al., High-resolution Fourier-transform 93. Y. Gao et al., Photon-trapping microstructures enable
37, 032002 (2019). doi: 10.1116/1.5085259 spectrometer chip with microphotonic silicon spiral high-speed high-efficiency silicon photodiodes. Nat. Photonics
48. P. J. Lapray, X. Wang, J. B. Thomas, P. Gouton, Multispectral waveguides. Opt. Lett. 38, 706–708 (2013). doi: 10.1364/ 11, 301–308 (2017). doi: 10.1038/nphoton.2017.37
filter arrays: Recent advances and practical implementation. OL.38.000706; pmid: 23455272 94. B. Craig, V. R. Shrestha, J. Meng, J. J. Cadusch, K. B. Crozier,
Sensors 14, 21626–21659 (2014). doi: 10.3390/s141121626; 71. H. Podmore et al., Demonstration of a compressive-sensing Experimental demonstration of infrared spectral reconstruction
pmid: 25407904 Fourier-transform on-chip spectrometer. Opt. Lett. 42, using plasmonic metasurfaces. Opt. Lett. 43, 4481–4484
49. J. H. Correia, G. de Graaf, S. H. Kong, M. Bartek, 1440–1443 (2017). doi: 10.1364/OL.42.001440; (2018). doi: 10.1364/OL.43.004481; pmid: 30211895
R. F. Wolffenbuttel, Single-chip CMOS optical pmid: 28362792 95. J. Meng, J. J. Cadusch, K. B. Crozier, Detector-only
microspectrometer. Sens. Actuators A 82, 191–197 (2000). 72. J. Li, D. F. Lu, Z. M. Qi, Miniature Fourier transform spectrometer based on structurally colored silicon nanowires
doi: 10.1016/S0924-4247(99)00369-6 spectrometer based on wavelength dependence of half-wave and a reconstruction algorithm. Nano Lett. 20, 320–328
50. S.-H. Kong, J. H. Correia, G. de Graaf, M. Bartek, voltage of a LiNbO3 waveguide interferometer. Opt. Lett. 39, (2020). doi: 10.1021/acs.nanolett.9b03862; pmid: 31829611
R. F. Wolffenbuttel, Integrated silicon microspectrometers. 3923–3926 (2014). doi: 10.1364/OL.39.003923; 96. B. Zheng et al., On-chip measurement of photoluminescence
IEEE Instrum. Meas. Mag. 4, 34–38 (2001). doi: 10.1109/ pmid: 24978772 with high sensitivity monolithic spectrometer. Adv. Opt.
5289.953457 73. M. C. M. M. Souza, A. Grieco, N. C. Frateschi, Y. Fainman, Mater. 8, 2000191 (2020). doi: 10.1002/adom.202000191
51. S. W. Wang et al., Concept of a high-resolution miniature Fourier transform spectrometer on silicon with thermo-optic 97. J. Dong, The smallest nanowire spectrometers.
spectrometer using an integrated filter array. Opt. Lett. 32, non-linearity and dispersion correction. Nat. Commun. 9, Front Optoelectron. 12, 341 (2020). doi: 10.1007/
632–634 (2007). doi: 10.1364/OL.32.000632; 665 (2018). doi: 10.1038/s41467-018-03004-6; s12200-019-0983-5
pmid: 17308584 pmid: 29445152 98. M.-N. Zhang et al., Spectrum projection with a bandgap-gradient
52. N. K. Pervez et al., Photonic crystal spectrometer. Opt. 74. S. N. Zheng et al., Microring resonator-assisted Fourier perovskite cell for colour perception. Light Sci. Appl. 9, 162
Express 18, 8277–8285 (2010). doi: 10.1364/OE.18.008277; transform spectrometer with enhanced resolution and large (2020). doi: 10.1038/s41377-020-00400-w; pmid: 33014357
pmid: 20588673 bandwidth in single chip solution. Nat. Commun. 10, 2349 99. H. Sun et al., In Situ Formed Gradient Bandgap-Tunable
53. Q. Hang, B. Ung, I. Syed, N. Guo, M. Skorobogatiy, Photonic (2019). doi: 10.1038/s41467-019-10282-1; pmid: 31138800 Perovskite for Ultrahigh-Speed Color/Spectrum-Sensitive
bandgap fiber bundle spectrometer. Appl. Opt. 49, 75. D. M. Kita et al., High-performance and scalable on-chip Photodetectors via Electron-Donor Control. Adv. Mater. 32,
4791–4800 (2010). doi: 10.1364/AO.49.004791; digital Fourier transform spectroscopy. Nat. Commun. 9, e1908108 (2020). doi: 10.1002/adma.201908108;
pmid: 20820222 4405 (2018). doi: 10.1038/s41467-018-06773-2; pmid: 32080927
54. A. Tittl et al., Imaging-based molecular barcoding with pmid: 30353014 100. C. C. Chang, H. N. Lee, On the estimation of target spectrum
pixelated dielectric metasurfaces. Science 360, 1105–1109 76. A. Herrero-Bermello et al., On-chip Fourier-transform for filter-array based spectrometers. Opt. Express 16,
(2018). doi: 10.1126/science.aas9768; pmid: 29880685 spectrometers and machine learning: A new route to smart 1056–1061 (2008). doi: 10.1364/OE.16.001056;
55. A. Thelen, Circularly wedged optical coatings I: Theory. Appl. photonic sensors. Opt. Lett. 44, 5840–5843 (2019). pmid: 18542178
Opt. 4, 977 (1965). doi: 10.1364/AO.4.000977 doi: 10.1364/OL.44.005840; pmid: 31774793 101. P. C. Hansen, Rank-Deficient and Discrete Ill-Posed Problems
56. A. M. Mika, Linear-wedge spectrometer. Proc. SPIE 1298, 77. E. le Coarer et al., Wavelength-scale stationary-wave (SIAM, 2005), vol. 4.
127–131 (1990). doi: 10.1117/12.21343 integrated Fourier-transform spectrometry. Nat. Photonics 1, 102. C.-C. Chang, H.-Y. Lin, Spectrum reconstruction for on-chip
57. M. Grundmann, Modeling of a waveguide‐based UV-VIS-IR 473–478 (2007). doi: 10.1038/nphoton.2007.138 spectrum sensor array using a novel blind nonuniformity
spectrometer based on a lateral (In,Ga)N alloy gradient. 78. X. Nie, E. Ryckeboer, G. Roelkens, R. Baets, CMOS- correction method. IEEE Sens. J. 12, 2586–2592 (2012).
Phys. Status Solidi A 216, 1900170 (2019). doi: 10.1002/ compatible broadband co-propagative stationary Fourier doi: 10.1109/JSEN.2012.2197609
pssa.201900170 transform spectrometer integrated on a silicon nitride 103. Z. Wang, Z. Yu, Spectral analysis based on compressive
58. W. A. Hovis Jr.., W. A. Kley, M. G. Strange, Filter wedge photonics platform. Opt. Express 25, A409–A418 (2017). sensing in nanophotonic structures. Opt. Express 22,
spectrometer for field use. Appl. Opt. 6, 1057–1058 (1967). doi: 10.1364/OE.25.00A409; pmid: 28437970 25608–25614 (2014). doi: 10.1364/OE.22.025608;
doi: 10.1364/AO.6.001057; pmid: 20062123 79. D. Pohl et al., An integrated broadband spectrometer on pmid: 25401594
59. O. Pust, Innovative filter solutions for hyperspectral imaging. thin-film lithium niobate. Nat. Photonics 14, 24–29 (2020). 104. S. Zhang, Y. Dong, H. Fu, S. L. Huang, L. Zhang, A spectral
Optik Photon. 11, 24–27 (2016). doi: 10.1002/ doi: 10.1038/s41566-019-0529-9 reconstruction algorithm of miniature spectrometer
opph.201600012 80. B. Redding, S. M. Popoff, H. Cao, All-fiber spectrometer based on sparse optimization and dictionary learning.
60. A. Emadi, H. Wu, G. de Graaf, R. Wolffenbuttel, Design and based on speckle pattern reconstruction. Opt. Express 21, Sensors 18, 644 (2018). doi: 10.3390/s18020644;
implementation of a sub-nm resolution microspectrometer 6584–6600 (2013). doi: 10.1364/OE.21.006584; pmid: pmid: 29470406
based on a Linear-Variable Optical Filter. Opt. Express 20, 23482230 105. Y. M. Eltagoury, Y. M. Sabry, D. A. Khalil, All-silicon double-
489–507 (2012). doi: 10.1364/OE.20.000489; pmid: 22274371 81. B. Redding, S. F. Liew, R. Sarma, H. Cao, Compact cavity Fourier-transform infrared spectrometer on-chip.
61. R. G. DeCorby, N. Ponnampalam, E. Epp, T. Allen, spectrometer based on a disordered photonic chip. Nat. Adv. Mater. Technol. 4, 1900441 (2019). doi: 10.1002/
J. N. McMullin, Chip-scale spectrometry based on tapered Photonics 7, 746–751 (2013). doi: 10.1038/nphoton.2013.190 admt.201900441

Yang et al., Science 371, eabe0722 (2021) 29 January 2021 10 of 11

RES EARCH | R E V I E W

106. R. Mannila et al., Gas detection with microelectromechanical learning. Nat. Commun. 10, 2903 (2019). doi: 10.1038/ 116. N. Hagen, M. W. Kudenov, Review of snapshot spectral
Fabry-Perot interferometer technology in cell phone. s41467-019-10827-4; pmid: 31263102 imaging technologies. Opt. Eng. 52, 090901 (2013).
Proc. SPIE 9482, 94820P (2015). doi: 10.1117/12.2176923 111. NeoSpectra-micro FTIR (2016); www.neospectra.com/our- doi: 10.1117/1.OE.52.9.090901
107. J. H. Correia, M. Bartek, R. F. Wolffenbuttel, High-selectivity offerings/neospectra-micro/. 117. X. Cai et al., One-shot ultraspectral imaging with reconfigurable
single-chip spectrometer in silicon for operation at visible 112. P. Kotidis et al., Optical, tunable filter-based micro- metasurfaces. arXiv 2005.02689 [physics.optics] (7 May 2020).
part of the spectrum. IEEE Trans. Electron Dev. 47, 553–559 instrumentation for industrial applications. Proc. Instr. Syst. 118. K. Monakhova, K. Yanny, N. Aggarwal, L. Waller, Spectral
(2000). doi: 10.1109/16.824727 and Autom. Soc. (ISA), (2003). www.thevespiary.org/library/ DiffuserCam: Lensless snapshot hyperspectral imaging with a
108. A. Herrero-Bermello et al., Temperature dependence Files_Uploaded_by_Users/no1uno/pdf/Instrumentation/IR/ spectral filter array. Optica 7, 1298 (2020). doi: 10.1364/
mitigation in stationary Fourier-transform on-chip Kotidis.etal.Optical.Tunable.Filter.Based.Micro. OPTICA.397214
spectrometers. Opt. Lett. 42, 2239–2242 (2017). Instrumentation.for.Industrial.Applications.pdf.
doi: 10.1364/OL.42.002239; pmid: 28569891 113. Changhong H2 spectrometer phone (2017); www.cnet.com/
109. D. Jha et al., Enhancing materials property prediction pictures/changhong-h2-ces-2017/3/. AC KNOWLED GME NTS
by leveraging computational and experimental data 114. Electronic device comprising plurality of light sources. U.S. We thank W. Han for valuable discussions. Funding: Supported
using deep transfer learning. Nat. Commun. 10, 5316 Patent 20190290148A1 (2018). by the National Natural Science Foundation of China (51706141,
(2019). doi: 10.1038/s41467-019-13297-w; 115. A. Koppelhuber, O. Bimber, Towards a transparent, flexible, 51976122) and EPSRC (EP/L016087/1, EP/T014601/1).
pmid: 31757948 scalable and disposable image sensor using thin-film Competing interests: The authors declare no competing interests.
110. J. S. Smith et al., Approaching coupled cluster accuracy with luminescent concentrators. Opt. Express 21, 4796–4810
a general-purpose neural network potential through transfer (2013). doi: 10.1364/OE.21.004796; pmid: 23482014 10.1126/science.abe0722

Downloaded from https://www.science.org on December 29, 2023

Yang et al., Science 371, eabe0722 (2021) 29 January 2021 11 of 11

 Miniaturization of optical spectrometers
Zongyin Yang, Tom Albrow-Owen, Weiwei Cai, and Tawfique Hasan

Science 371 (6528), eabe0722. DOI: 10.1126/science.abe0722

Miniaturizing spectrometers
Optical spectroscopy is a widely used characterization tool in industrial and research laboratory settings for chemical
fingerprinting and analysis. High-end spectrometers are typically benchtop based with bulky optical components,
moving parts, and long path lengths, and they can deliver a wealth of information with ultrahigh precision and
bandwidth. There is, however, a drive toward miniaturization of spectrometers, in which concepts in nanophotonics
are used to control light on much smaller scales. Yang et al. reviewed recent developments in spectrometry systems,

Downloaded from https://www.science.org on December 29, 2023

including various fabrication approaches of nanophotonics systems and the software that computationally determines
the spectra, that strive to shrink their footprint and open up applications in portable spectroscopy.
Science, this issue p. eabe0722

View the article online

https://www.science.org/doi/10.1126/science.abe0722
Permissions
https://www.science.org/help/reprints-and-permissions

Use of this article is subject to the Terms of service

Science (ISSN 1095-9203) is published by the American Association for the Advancement of Science. 1200 New York Avenue NW,
Washington, DC 20005. The title Science is a registered trademark of AAAS.

Copyright © 2021 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim
to original U.S. Government Works