Low Loss Coupling using Silicon Grating Couplers

Allan Zhang
April 2019

1 Introduction
The inherent limitation of the electronic circuits has gradually revealed itself as the expo-
nential growth rate of the computational capacity afforded by VLSI chips as predicted by
Moore’s Law comes to a halt. Capacitance and inductance fundamentally limits the speed of
electrons, which in turn limits the bandwidth and latency of electronic circuits. Also Ohm’s
aw dictates that as transistor and interconnect density increases, the amount of heat to be
dissipated also increases, to the point where cooling by air is no longer possible. As one of
the promising solutions to the drawbacks of electronic circuits, people have devised methods
to utilize light as information carrier instead. Using photons to carry information is indeed
revolutionary since photons travel many orders of magnitude faster than electrons 108 m/s
compared to 105 –106 m/s [1], and incurs minimal energy loss in transparent media. More
importantly, optical channels do not suffer from EM radiation, ie: wave interference. This
allows multiple pathways to coexist within the same optical component, eg: multi-mode op-
tical fibers, significantly increasing the bandwidth of data transfers. Fiber optics and silicon
photonics have thus emerged as new fields of study having the potential to augment or even
replace conventional electronics. Fiber optics study the application of optical fibers, long
glass fibers consisting of a higher index core and lower index cladding materials. Light is
confined within the core by means of total internal reflection and thereby transmits along
the core. Conventional fiber optics have enabled the long-range, low-latency communications
worldwide. However, in recent years sub-meter range computer communication via optical
link has been possible, allowing for much faster data transfers as required by the increase
in computational capacity. Now research is underway to further shrink the dimension of
optical channels, ie: nanometer scale so that optical pathways between distinct chips and
even between individual components within a chip are possible. This is enabled by the field
of silicon nanophotonics. Silicon makes a great candidate for the manipulation of light due
to its nonlinear optical properties, eg: the Kerr effect, wave mixing, etc., allowing light to
interact with light. Silicon also has high reflective index, granting it its waveguiding prop-
erties. Moreover, silicon has already been conveniently used as the substrate for integrated
circuits. Thus, existing nanofabrication technologies for ICs can also be used to fabricate
silicon photonic devices. As of now , silicon photonic devices that have been or are being
developed include the high speed Ge-on-Si photodetectors [1], low loss waveguides [2], ulta-
high Q optical resonators [3], AND and NAND logic [4], and electro-optical modulators [5],
and Si Raman Lasers[6]. In December 2015 the first microprocessor with optical I/O was

1
demonstrated implementing the ”zero-change” CMOS technology [7]. The future for optical
circuits enabled by silicon photonics is indeed promising. Former Intel senior advisor Pat
Gelsinger is quoted as saying that ”Today, optics is a niche technology. Tomorrow, it’s the
mainstream of every chip that we build.” [8]

2 Challenge and Objective

Silicon waveguides are the essential building blocks of optical ICs. Thanks to the high refrac-
tive index of silicon, silicon waveguides can be extremely compact compared to conventional
waveguides such as the optical fiber. Typical silicon photonic waveguides are 500nm by
200nm compared to tens of micrometers in diameter for optical fibers. This inherent mis-
match in size presents serious challenges for coupling off-chip optical signals, eg: from fiber
optic systems, into sub-micrometer scale nanophotonic devices. Typically, single-mode opti-
cal fibers are 10.4 micrometers in diameter. In comparison, a typical silicon strip waveguide
has a core diameter of only 100-300 nm. [9]

Figure 1: Size comparison between free space mode and waveguide mode

The direct consequence of this size mismatch is the low efficiency of the system. Assume
the core diameter of silicon waveguide is 200 nm, then
Ewaveguide
Percentage of energy lost = 1 −
Ef iber
Areawaveguide
=1−
Areaf iber
Radiuswaveguide 2
=1−( )
Radiusf iber
200nm 2
=1−( )
10400nm
= 99.96%

2
That is, only 0.04% of energy is transmitted into the waveguide, resulting in an unacceptable
amount of energy lost. Moreover, the difference in NA (numerical aperture) contributes to
the poor efficiency. Lastly, due to the substantial energy loss, when light is coupled out
from waveguide mode back into fiber mode, the signal strength would be so low that signal
recognition would be impossible.
This issue would not have concerned us if we could manage to generate on-chip optical
signals. In fact silicon is already being used as light sources in devices such as Si Raman Laser,
Brillouin Laser and hybrid silicon lasers. [10] However, the technology of such devices are far
from mature. The challenge of generating photons from silicon lies in the fact photon emission
is much more difficult since silicon’s indirect band gap requires not only that the absorbed
photon satisfies the energy gap but also that the the absorbed photon and emitted photon
match the phase difference. Therefore, in many cases coupling between fiber/waveguide
mode is still common and very much relevant.

3 Survey of diff coupling methods

3.1 Lensed Fiber
Many solutions to the above-mentioned size mismatch between fiber and waveguide mode
have been proposed. In investigating the effectiveness of such solutions, one must understand
that our objective lies not only in the successful coupling of light between the two modes
but also that the solution attains a satisfactory coupling efficiency. After all, if there is only
slight improvement in efficiency, even if signal recovery from the chip is possible the amount
of energy loss is still unacceptable.
The most direct solution is to focus the light from the optical fiber to a scale that is
acceptable for coupling into the waveguide. The device that accomplishes this is called
lensed fiber. [11] The lensed fiber is a modified version of the conventional fiber, adding to
the tip of which a lens-like aperture. Much like a converging lens would converge light, the
lens fiber converges light at a focal spot some distance beyond the fiber edge. The focal length
could be manipulated so that the focal spot falls exactly on the entrance of the waveguide so
that light could be coupled in. Nanofabrication technologies, such as the nanoimprint and

Figure 2: SEM image of a lensed fiber with nanoimprinted lens out of high refractive index
material

the electron beam lithography, allow us extremely precise control over the shape, aperture
size, and focal power of the lensed fiber. It is shown that with precise alignment and correct

3
selection of focal length, the fiber-to-fibber loss, that is, the total loss from the input fiber,
in through the waveguide and out to the output fiber, can be as low as 20db. This number
is still far from satisfactory.

Figure 3: Beam profile comparison before and after focusing. Focal spot diameter is 810 nm
@ 1/e2 for 660 nm wavelength.

3.2 Overcladding and Inverse Taper

Other solutions seek to alter the geometry or the refractive index profile of the optical fiber.
One such example is called overcladding [11], a technique to clad the core with multiple
claddings so that the desired refractive index profile can be achieved. The practice of over-
cladding serves two purposes. Firstly, overcladding increases the diameter of the preform,
the precursor to optical fiber in the manufacturing process, so that the required thickness can
be reached. This facilitates the subsequent extension of the precursor to desired length and
diameter, called the drawing process. Secondly, overcladding adds to the original cladding a
second layer of cladding with a different refractive index, thereby creating a gradient refrac-
tive index. This refractive index profile of the fiber can thus be altered to insure that the
light travelling within the fiber never reaches the critical angle of reflection, which reduces
attenuation. This helps to keep the wave profile concentrated and facilitates coupling.

Figure 4: Crosssection scheme of the standard optical fiber

Other methods that have been proposed include the multi-dielectric structures [12] which
utilizes a stepwise parabolic graded index profile combined with a horizontal taper that
confine light in both forwards and backwards directions, and the inverted lateral taper [13]
with a polymer overlay that was able to achieve 1db loss. Although such solutions are

4
 (a) Schematic of the inverse taper coupler

(b) Schematic configuration of the present mode

converter based on an inverse taper for multimode
silicon photonic integrated circuits

Figure 5: The inverse taper coupler

extremely efficient, they have limited application due to various pre-existing conditions that
need to be satisfied before the methods can be applied. The multi-dielectric method applies
only to high-refractive-index-contrast systems while the inverse taper method requires that
it must be implemented and on the edge of the chip. Moreover, such methods often greatly
increases high production complexity, such as chip cleaving and facet polishing. The extra
manufacturing requirements prevents such setups from mass production.

4 Principle of Operation of Grating Coupler

4.1 Device Setup
Diffraction tells us that when a wavefront is incident on diffracting elements, simply blocking
objects, each element acts as a new source of the wavefront and that those waves propagate
beyond the diffracting elements and interfere to form diffraction patterns. An important
application of diffraction is when the diffracting elements are periodically positioned such
that the the amplitude/phase of the incident wave is periodically modified. Objects that
include the foregoing setup is called the diffraction grating.
Grating couplers that couple light into thin film waveguide have long been known.[14]
The grating coupler diffracts light much like the Bragg diffraction. The bragg diffraction
occurs for when X ray is incident on the crystal lattice. The diffraction grating is analogous
to a 1-D crystal lattice since it is nothing more than a periodic distribution of ”atoms” to
incoming light. The bragg’s law tells us that for lattice separation d, incident angle θ and
incoming wavelength λ constructive interference occurs when

2d sin θ = nλ

5
The following illustration is a simplified version of the diffraction grating coupler setup. The
input light from free space mode (fiber mode) is incident on the gratings at an angle. Then
the extra-planar input light is coupled into planar light and propagates to the waveguide.
Similarly, waveguide mode may couple out into fiber mode through another grating coupler.

Figure 6: Illustration of grating coupling between optical fibers and an SOI sample

In the illustration, the silicon photonic waveguide consists of silicon on top of a layer of
silica (silicon dioxide). The is also true for most standard SOI (silicon on insulator) silicon
strip waveguides.[17] On a real chip, the input fiber would be the free space light source
and the output fiber the free space detector. The space in the illustration between the two
grating couplers would normally consist of two taper devices to focus light and other photonic
devices found on a chip.

4.2 Theory of Operation

Each diffraction grating is identified by a grating vector. The grating vector K describes
the periodicity of the gratings and is defined as a vector with magnitude |K| = 2πΛ, where
Λ is the grating period, and is pointed along the direction of periodicity. The relationship
between the diffracted wave vector and incident wave vector is such that the former is the
vector addition of the undiffracted wave vector and the scalar multiple of the grating vector
as specified in the following equation [19]

Km = Ku − mK (1)

, where m ∈ Z is the diffracted order number, Ku is the wave vector of the undiffracted
incident wave, that is, Ku = K0 , and Km is the wave vector of the mth diffracted wave.
The equation that relates the wave vector of the diffracted to the incident wave in a
periodic structure is called the FPBC (Floquet periodic boundary conditions). Floquet
waves is the set of all resulting diffracted waves from a periodic structure as dictated by the
FPBC. [14] It is important to note that despite the fact that the Floquet waves is an infinite
set of waves corresponding to an infinite number of integer diffraction orders m, there exists
only a limited number of such Floquet waves that is physically possible. To see this, let
us decompose the wave vectors in equation 1 to heir respective in-plane x̂ and out-of-plane
vector ẑ components, namely

6
 Figure 7: Geometry of two dimensional surface relief grating

2πnt 2πnu 2πm

(sinθm x̂ + cos θm ẑ) = (sinθu x̂ + cos θu ) − x̂ (2)
λ λ Λ
where nt is the refractive index of the diffraction medium and θu is the angle of the incident
wave, θm is the angle of diffracted wave of mth order, λ is the wavelength of incident light
and Λ is the periodicity of diffraction gratings. We examine the in-plane component of the
equation, namely
λ
nt sinθm = nu sinθu − m (3)
Λ
and solve for the incident angle θ where resonance occurs.
Finally we relate the diffracted and undiffracted light by the Snell’s Law and substitute
ni for nu and θi for θu and
λ
nt sinθm = ni sinθi − m (4)
Λ
Note that equation 4 reduces simply to Snell’s Law for m = 0, or when there is no diffraction
grating, ie: smooth interface between incident medium and transmitted medium.
Equation 4 describes the principal of operation for the diffraction grating couplers. From
equation 4 we see that diffraction angle θm depends on both the incident wavelength and the
periodicity of the diffraction gratings. If we choose periodicity Λ small enough compared to λ
we can restrict dependence of theta solely on wavelength. Now what we have is a dispersion
relationship between the wavelength and the diffraction angle. The fact that the diffraction
grating is dispersive lends itself perfectly to many optical applications. [16]

5 Grating Coupler Fabrication [20]

The central idea underlying the fabrication process is secondary lithography [17] that involves
a separate lithography for the waveguide and the diffraction gratings. The initial wafer is
prepared of coating in Step 1. In step 2, a negative photoresist coating is applied to the

7
Figure 8: Step-by-step illustration of grating coupler fabrication. Photo credit: see reference
[11]

silicon-on-silica substrate. Negative photoresist is chosen since it more time and energy
efficient to expose only the grating patterns compared to their complement. Now in step 3
electron beam lithography expose the PR to develop the desired waveguide patterns on the
PR. Then in the step 4 the patterns are transferred to the silicon layer by means of reactive
ion etch (RIE). Next, in step 5 developer solution is applied to the PR coating to remove
it. In step 6, we again coat the waveguide with negative PR and similar to step 3 in step 7
we apply secondary electron beam lithography to develop patterns for the gratings on PR.
Next, in step 8 the patterns are RIE etched onto the silicon layer. After removing the PR
coating in step 9, we complete the fabrication process for the diffraction gratings coupler on
top of the waveguide.

8
References
[1] R. W. Keyes ”Physical limits of silicon transistors and circuits” IBM Research Divi-
sion, Yorktown, NY 10598, USA Received 27 April 2005, in final form 11 August 2005
Published 19 September 2005 [Online] at stacks.iop.org/RoPP/68/2701

[2] S. Xiao et al., ”Compact silicon microring resonators with ultra-low propagation loss in
the C band,” Optics Express, vol. 15, no. 22, pp. 14467-14475, Oct 2007.

[3] P. Dong al., ”Low loss shallow-ridge silicon waveguides,” Optics Express, vol. 18, no.
14, pp. 14474-14479, Jul 2010.

[4] F. Xia, L. Sekaric, and Y. Vlasov, ”Ultracompact optical buffers on a silicon chip,”
Nature Photonics, vol. 1, pp. 65-71, 2007.

[5] Q. Xu and M. Lipson, ”All-optical logic based silicon micro-rong resonators,” Optics
Express, vol. 15, no. 3, pp. 924-929, Feb 2007.

[6] G. T. Reed et al., ”Silicon optical modulators,” Nature Photonics, vol. 4, pp. 518- 526,
2010.

[7] Sun, Chen; et al. (2015). ”Single-chip microprocessor that communicates directly using
light”. Nature. 528: 534–538. Bibcode:2015Natur.528..534S. doi:10.1038/nature16454.

[8] ”Silicon Photonics”. Intel. Retrieved 14 July 2009.

[9] D. TAILLAERT, F. V. LAERE, M. AYRE1, W. BOGAERTS, D. V. THOURHOUT, P.

BIENSTMAN and R. BAETS ”Grating Couplers for Coupling between Optical Fibers
and Nanophotonic Waveguides” Japanese Journal of Applied Physics Vol. 45, No. 8A,
2006, pp. 6071–6077

[10] M. A. Foster et al., ”Broad-band optical parametric gain on a silicon photonic chip,”
Nature, vol. 441, pp. 960-963, 2006.

[11] aBeam Technologies ”Lensed fiber with high refractive index lens” [Online].
https://fiberphotonics.com/products/lensed-fiber/

[12] T. Tsuchizawa et al., ”Microphotonics devices based on silicon microfabrication tech-

nology,” IEEE Journal of Selected Topics in Quantum Electronics, vol. 11, no. 1, pp.
232-240, Jan-Feb 2005.

[13] V. Nguyen et al., ”Silicon-based highly-efficient fiber-to-waveguide coupler for high index
contrast systems,” Applied Physics Letters, vol. 88, Feb 2006.

[14] V. R. Almeida et al., ”Nanotaper for compact mode conversion,” Optics Letter, vol. 28,
no. 15, pp. 1302-1304, Aug 2003.

[15] R. Orobtchouk, A. Layadu, H. Gualous, D. Pascal, A. Koster and S. Laval: Appl. Opt.
39 (2000) 5773.

9
[16] D. TAILLAERT, F. V. LAERE, M. AYRE1, W. BOGAERTS, D. V. THOURHOUT, P.
BIENSTMAN and R. BAETS ”Grating Couplers for Coupling between Optical Fibers
and Nanophotonic Waveguides” Japanese Journal of Applied Physics Vol. 45, No. 8A,
2006, pp. 6071–6077

[17] Graham T. Reed, Silicon photonics: the state of the art, 1st ed. West Sussex, UK:
Wiley-Interscience, 2008. structures

[18] Frei (2014), Modeling Electromagnetic Waves and Periodic Structures [Online]
https://www.comsol.com/blogs/modeling-electromagnetic-waves-periodic-structures/

[19] R. Chu and J. A. Kong, “Modal theory of spatially periodic media,” IEEE Trans.
Microwave Theory and Tech., vol. 25, pp. 18-24, Jan. 1997.

[20] K. R. Harper, ”Theory, Design, and Fabrication of Diffractive Grating

Coupler for Slab Waveguide” (2003). All Theses and Dissertations. 101.
https://scholarsarchive.byu.edu/etd/101

10