Nanotechnology

PAPER

High flow rate nanofluidics for in-liquid electron microscopy and

diffraction
To cite this article: Ariel A Petruk et al 2019 Nanotechnology 30 395703

View the article online for updates and enhancements.

This content was downloaded from IP address 35.176.47.6 on 21/08/2019 at 08:22

 Nanotechnology

Nanotechnology 30 (2019) 395703 (8pp) https://doi.org/10.1088/1361-6528/ab2cf2

High flow rate nanofluidics for in-liquid

electron microscopy and diffraction
Ariel A Petruk, Caroline Allen1, Nicolás Rivas, Kostyantyn Pichugin and
Germán Sciaini
The Ultrafast electron Imaging Lab (UeIL), Department of Chemistry and Waterloo Institute for
Nanotechnology (WIN), University of Waterloo, 200 University Ave. W., N2L 3G1, Waterloo, Ontario,
Canada

E-mail: gsciaini@uwaterloo.ca

Received 8 April 2019, revised 11 June 2019

Accepted for publication 26 June 2019
Published 15 July 2019

Abstract
We introduce a nanofluidic platform that can be used to carry out femtosecond electron
diffraction (FED) and transmission electron microscopy (TEM) measurements in liquid samples
or in-liquid specimens, respectively. The nanofluidic cell (NFC) system presented herein has
been designed to withstand high sample refreshing rates (over one kilohertz), a prerequisite to
succeed with FED experiments in our lab. Short beam paths, below 1 μm, in combination with
ultrathin membranes (less than 100 nm thick) are necessary conditions for in-liquid FED and
TEM studies due to the strongly interacting nature of electrons. Depending on the application,
the beam path in our NFC can be tuned between 50 nm and 10 μm with ultrathin stoichiometric
silicon nitride (Si3N4) windows as thin as 20 nm. Stoichiometric Si3N4 has been selected to
reduce membrane bulging owing to its higher tensile stress and transparency in the UV–vis–NIR
region to allow for laser excitation in FED experiments. Key design parameters and
improvements made over previous NFC systems are discussed, and some preliminary electron
images obtained by 200 kV scanning TEM are presented.

Supplementary material for this article is available online

Keywords: nanofluidics, in-liquid imaging, in-liquid TEM, time-resolved electron diffaction

(Some figures may appear in colour only in the online journal)

1. Introduction
The use of environmental cells for the study of hydrated
specimens in the transmission electron microscope (TEM)
dates back to 1944 [1]. Innovative nanofluidic cells (NFCs) or
nanofluidic cell systems (NFCSs) can literally transform any
standard TEM into an environmental TEM and beyond by
enabling the study of specimens in liquids. In terms of liquid
circulation capability, NFCs can be subdivided into two major
categories, ‘stationary’ and ‘flow’ cells. Stationary cells are
somewhat simpler in their design and, arguably, in usage but
inherently lack the possibility of refreshing the liquid
environment and/or the sample itself. Original stationary
1
Present address: EMSL Analytical, Inc. 2756 Slough St, L4T 1G3,
Mississauga, Ontario, Canada.
cells employed nitrocellulose [2], and later carbon films [3, 4]
to culminate with the use of graphene sheets [5] as a means to
encapsulate liquid. Despite the excellent electrical and ther-
mal conductivities, and the electron beam transparency of
graphene, silicon nitride (SiNx) has become the most popular
window material for in-liquid TEM and a variety of appli-
cations [6].
Several pioneering in situ TEM experiments have been
performed with stationary SiNx-NFCs and provided great
insights into reactions and processes occurring in liquids from
a molecular level of inspection. These include nanoparticle
(NP) nucleation and growth [7–13], NP assembly [14, 15] NP
etching [16], NP motion [17–19], NP uptake by cells [20], NP
solvation layers [21], and reactions in living organisms [22].
The main advantages of SiNx are its compatibility with many
Si-device fabrication techniques and its good mechanical and

0957-4484/19/395703+08$33.00 1 © 2019 IOP Publishing Ltd Printed in the UK

Nanotechnology 30 (2019) 395703 A A Petruk et al

chemical properties [23, 24]. Low-pressure chemical vapor

deposition (LPCVD) is commonly used to fabricate free-
standing SiNx windows in Si wafers, which after sectioning,
produce a large number of NFC chips. LPCVD SiNx is
amorphous, composed of tetrahedral SiN4 and trigonal NSi3
units connected without long-range order [23]. The ratio of Si
to N can be varied ranging from Si-poor SiNx to stoichio-
metric silicon nitride (Si3N4) to Si-rich SiNx films. Si-rich
nitride is referred to as low-stress SiNx owing to its low
residual stress that makes it more flexible and less prone to
fracture than Si3N4. To form an NFC, the windows have to
spatially overlap in order to grant a sufficient viewing area for
electron diffraction and electron imaging experiments to be
performed in transmission geometry. Thus, rectangular win-
dows in a cross orientation are commonly implemented to
facilitate window–window overlap during cell assembly.
Window thicknesses (Wt) are typically in range of 10 nm to
100 nm. Chips are usually separated by a spacer with a
thickness (SPt) that is normally between 100 nm and 5 μm.
The NFCs must be sealed against the body of the holder to
preserve the vacuum integrity of TEMs or femtosecond
electron diffractometers (FEDs) [25]. The spacer is one of the
most variable design features (see [26] for more details). SPt
must be appropriate for the desired sample and type of study; Figure 1. Simplified schematics of various ‘flow’ NFCs in a holder
(not at scale). Left, front cross-sectional views; the liquid flows in the
for observation of biological cells or other large structures, for direction perpendicular to the plane of the page. Right, lateral cross-
instance, the spacer should be thick enough to accommodate sectional views; the liquid flows from left to right.
the specimen height if squeezing is not desirable [27]. For
small in-liquid samples such as NPs, SPt can be minimized to
reduce the thickness of the liquid layer, and therefore enhance goes around the NFC sandwich; i.e. between the outer surface
the spatial resolution. of the NFC sandwich and the inner walls of the containing
Similar experiments to those mentioned above for static holder. This is a well known issue that companies are trying
cells have been successfully performed with flow cells to solve through the application of custom shaped O-rings
[27–37]. Flow NFCs should in principle confer better control [38], but unfortunately this is only one part of a more complex
over the effective thickness of the fluid layer (FLt). These are problem (vide infra). Figure 1(b) illustrates an improvement
often connected to a syringe pump that allows to increase or of the system presented in panel (a); here the channel shape
decrease FLt by pushing or sucking fluid. NFC designs are along the direction of the liquid flow has been modified to
conceptually simple, however, achieving ‘true net flow’ in
reduce its effective length (ECL), and therefore the flow
such restrictive geometries poses a real challenge as a con-
resistance (figure 2). However, there is still a large difference
sequence of the small Reynolds numbers and the reduction of
between the two lateral cross-sections (channel versus sur-
inertial effects caused by large surface to volume ratios in
roundings), and therefore no definite control of the flow rate
nanochannels. Figure 1 exhibits various NFCs that have been
through the viewing area. The design displayed in figure 1(c)
attempted with different degrees of success in enabling liquid
incorporates an inlet, an outlet, and an enclosing spacer to
flow. Light blue depicts the region occupied by the fluid. The
height of the channels has been exaggerated for clarity. The force the liquid to flow pass the viewing window. A channel
simplest design, with two flat Si-chips, is depicted in length of about a couple of millimeters is, however, too long
figure 1(a). In this NFCS the resistance for liquid to flow pass (flow resistance too high) to reach our desired flow rate
the window is huge; here the liquid needs to be pushed for FED experiments, i.e. a longitudinal speed >10 mm s–1
through a channel with a miniature cross-section for a couple (e.g. >1 kHz refreshing rate for a 10 μm window). We would
of millimeters before it can reach the viewing window. A like to mention that the liquid pressure in the inlet cannot be
typical recommendation from suppliers is to start by drop increased arbitrarily. A brute force approach for supplying
casting one of the Si-dies. This process, however, may lead to liquid at higher pressure leads to significant bulging, die–die
large variations in the actual FLt if care is not taken to ensure separation and/or the rupture of the SiNx membranes (vide
the liquid does not invade the spacer area upon closing the infra section ‘preliminary work’). The NFC concept shown in
NFC sandwich (vide infra section ‘preliminary work’). This figure 1(d) combines both key features of the designs dis-
limits the amount of dispensed liquid volume to the nl level. played in panels (b) and (c). It is desirable to have a channel
Although capillary forces may help the liquid to initially flow, with a well-defined path connecting one or multiple inlets
there is no mechanism in such a design to guarantee a defined with one outlet over the viewing area, with the minimum
flow path. The expected outcome is that most of the liquid possible ECL, such as the channel shape shown in figure 1(d).

2
Nanotechnology 30 (2019) 395703 A A Petruk et al

Figure 2. (a) Sketch of the cross-sectional channel shape with reduced effective channel length (ECL). SPt and FLt denote the spacer and
liquid layer thicknesses, respectively. In an ideal situation, SPt=FLt. In the present case, the spacer has been etched in the top wafer. This
provides a cleaner surface and a wider range (from tens of nm to several μm) of possible spacer thicknesses than those achievable by metal
evaporation [39]. (b) Flow simulation performed in COMSOL Multiphysics [40] for creeping flow conditions (see text and supplementary
information available online at stacks.iop.org/NANO/30/395703/mmedia for more details). The flow rate was 0.3 μl min−1. This value was
set to achieve a refreshing rate of one kHz over an ECL=50 μm, with a channel width of 100 μm, and SPt=1 μm.

Over the past five years our team, at The Ultrafast elec-
tron Imaging Lab (UeIL), has tested different NFC design
concepts. We found that achieving the aforementioned flow
velocity was not possible with any of the NFCs shown in
figure 1, and therefore decided to perform flow simulations as
well as the nanofabrication of a new optimized NFC system
targeted to our specific needs. Upper estimates for the win-
dow deformation and some preliminary TEM results indicate
that our NFC design will soon enable FED experiments and
fast mixing during in-liquid TEM measurements.
2. Technical aspects of NFC designs
2.1. Flow simulations
The Microfluidics Module v5.3 from COMSOL Multiphysics
software [40] was used to simulate liquid flow. A tetrahedral
mesh was implemented, and water was selected as the fluid.
The simulated system replicates the channel shape shown in
figure 2(a), which is an enlarged view of the lateral channel
shape introduced in figure 1, panels (b) and (d).
We have performed a vast number of liquid flow simu-
lations at different flow rates with values of SPt and ECL in
the ranges of 0.5–5 μm and 40–400 μm, respectively. For
details about flow simulations please refer to [26]. The
goal was to obtain the values of pressure drop along ECL
(DPi – o =Pinlet − Poutlet) for different flow conditions in order
to estimate the pressure differential DP = (Pin - Pext ) acting
on the Si3N4 membranes. These served as parameters during
experimental window deformation measurements, vide infra
figure 3. Figure 2(b) shows a representative result from the
simulation of a fully developed liquid flow. For the simulation
the initial flow velocity was adjusted to confer a flow rate of
0.3 μl min−1, which our simulated channel geometry,
provided a longitudinal flow velocity of 50 mm s–1 through
the viewing window. The pressure at the outlet was set 0 and
creeping flow conditions, a simplification of laminar flow,
was implemented. The validity of this approximation was
confirmed by the small average Reynolds number∼0.1
obtained from our calculations.
We would like to stress the fact that our results for DPi – o
were found to be in excellent agreement (within 1.5% relative
error) with those obtained from the analytical equation
derived for a rectangular pipe [41] with a channel width (CW)
such that CW ? SPt,
12 r n ECL V
DPi--o = . (1 )
CW SPt 3
Here r =1 g cm–3 is the density of the liquid,
v(298 K)=0.892 mm2 s–1 is the kinematic viscosity, V=
0.30 μl min–1 is the volumetric flow rate. In our simulations
CW=100 μm, SPt=1 μm, and ECL=50 μm.
In an optimal channel geometry, ECL should govern DP
(or DPi – o ). As shown in equation (1), DPi – o presents a linear
dependence with ECL and V; i.e. for a given V, the shorter
ECL, the smaller DP and the membrane deformation are (vide
infra). We found that flow velocities of ∼10 mm s–1 are
excessive for STEM work. The flow must be stopped, or
specimens kept attached to the window(s) in order to achieve
a decent resolution <10 nm (vide infra). However, we believe
that by providing better control of the flow and minimal
window deformation, in situ TEM measurements involving
fluid mixing followed by fast stop-flow conditions will be
possible.
2.2. Fabrication of freestanding Si3N4 films and NFCs
To satisfy the requirements for FED experiments, NFC win-
dows were made of Si3N4 to allow for optical transparency in
the UV–vis–NIR spectral range. In addition, Si3N4 has a

3
Nanotechnology 30 (2019) 395703
Figure 3. Characteristic window deformation measurements as a function of differential pressure (DP =Pin−Pext). The experimental
procedure is outlined in panel (a). Measurements are shown for freestanding Si3N4 films with a thickness of 100 nm, and widths of about
100 μm and 20 μm, panels (b) and (c), respectively. The values of DP are given in the insets. The complete dataset of membrane deformation
measurements is presented in supplementary table 1. The nominal height resolution of the employed profilometer is 0.1 nm; however, we
determined from our measurements a resolution of 3 nm (rms).
higher tensile stress than low-stress SiNx membranes [42], a
fact that helps to minimize window bulging. Freestanding
Si3N4 windows with rectangular shapes of different widths
(ws) and thicknesses were fabricated in order to perform
membrane deformation measurements. This was a necessary
step to estimate the optimal window dimensions and desired
thickness before proceeding with the actual fabrication of
NFC chips, which required a large number of expensive
photomasks. Slit-shaped windows not only facilitate win-
dow–window overlap during NFC assembly. It is well known
[24] that for rectangular windows of length (wl ), width (ws ),
and wl /ws  4, the membrane deformation is governed by the
smallest dimension, i.e. ws. All fabrication processes were
carried out by UeIL team members at the Quantum NanoFab
core cleanroom facility of the University of Waterloo.
Detailed recipes and fabrication protocols can be found
elsewhere [26].
2.3. Window deformation measurements
We experienced that Si3N4 films with thicknesses below
15 nm break frequently during the nanofabrication process,
and therefore decided to develop NFCs with a minimum
window thickness of 20 nm. Window deformation measure-
ments were done for Si3N4 membranes with thicknesses of
100 nm, 50 nm and 25 nm, and ws of approximately 100 μm,
50 μm and 20 μm. The operational principle is shown in
figure 3(a); the tip of a contact profilometer (Veeco model
Dektak 150) scans the window width with a resolution step of
14 nm. The force employed on the tip was 0.3 mg. The
window is placed upside down in order to enable the tip of the
profilometer to reach the surface of the membrane. Pressur-
ized air was used to generate the desired value of DP. The gas
pressure was adjusted and calculated by measuring the change
in volume in a syringe and utilizing the ideal gas equation.
The values of maximum membrane deformation (h) were
obtained by subtracting a reference scan (unpressurized cell,
black circles) from each maximum (colored circles), as shown
in figure 3, panels (b) and (c). Note that the concave shape of
the reference scan trace (black) arises from the force exerted
by the tip of the contact profilometer used.
4
A A Petruk et al
The measured values of h as a function of ws, Wt, and
DP are listed in supplementary table 1. It should be men-
tioned that h is expected to represent an upper estimate of the
membrane deformation produced by a given DP arising from
actual liquid flow. Cohesive and adhesive liquid forces [41]
are expected to lead to smaller deformation. Nevertheless,
these measurements provided valuable information in a very
conservative manner. We concluded by combining our
simulation results and window deformation measurements,
that windows with ws =20 μm and Wt=25 nm would
deform to about 0.2×2=0.4 μm at our required flow rate
conditions, DP » 100 kPa (see supplementary information).
This amount of window bulging would add to SPt and
increase FLt accordingly. The factor of 2 comes from the fact
that there are two windows in an NFC. In our estimation, we
assumed that the outlet port has been connected to an evac-
uated container to bring Pout≈3 kPa (the vapor pressure of
water at room temperature), please refer to [39] for more
details. Note that an open outlet configuration would literally
increase Pout, and therefore Pin by 100 kPa (since DP is given
by the channel geometry and flow conditions). Moreover,
calculations for the tensile stress of our Si3N4 films obtained
from our window deformation measurements, see supple-
mentary information, were found in good agreement with
previous results [43].
3. Results and discussion
3.1. Preliminary work
Since only flow rates are normally reported in previous work
but detailed information about channel cross-sectional geo-
metry and ECL, we decided to fabricate the ‘best’
(figure 1(d)) among all designs shown in figure 1. We per-
formed measurements of optical beam path utilizing a solu-
tion of rhodamine-B of known concentration in ethanol to
monitor the NFC optical path as function of the liquid flow
rate. Ethanol was employed as the solvent in order to reduce
dimerization effects on the absorption spectra of rhodamine-
B. NFCs with ECL of about 1 mm and 300 μm, channel
Nanotechnology 30 (2019) 395703
lateral width (CW)=2 mm, and SPt =1 μm were tested. The
liquid flow rate was set at 10 μl min−1 in order to attain a
sample refreshing rate >1 kHz over a 50 μm window. We
found that dies get apart after a few minutes of operation
(approximately the time required for pressure buildup)
increasing FLt to >15 μm and reaching absorbance satur-
ation. The NFC sandwich is kept together through the use of
O-rings (vide supra figure 1), leading to chip–chip separation
when the overall internal force exceeds the external one
exerted by the compressed O-ring rubber. Si wafers used in
the fabrication of NFC chips have thicknesses in the range of
200–500 μm and are very fragile, and therefore O-rings are
not fully pressed into their grooves to avoid chip-to-holder
contact and die fracture.
We also noticed that once the liquid invades the spacer
region it was no longer possible to recover the initial FLt of
1 μm (i.e. spacer thickness SPt) by simply tightening the
screws to increase the compression force acting on the
O-rings. As a consequence of the enlarged cross-sectional
area, the flow velocity dramatically drops by about two orders
of magnitude, and with this the internal pressure (Pin) in the
NFC. Thus, FLt ? SPt and the refreshing rate becomes
insufficient for performing pump-probe experiments at 1 kHz.
The above observation is also important to the use static
NFCs requiring drop-casting (vide supra) since the liquid can
easily run on top of the spacer area, and therefore lead to
FLt ? SPt.
It should be mentioned that cleaning the surface of the
dies with piranha solution or oxygen plasma right before NFC
assembly and augmenting the initial compression of the
O-rings improved the flow conditions, but it did not solve the
problem. Besides piranha and oxygen plasma cleaning no
other surface treatments were explored. Die–die bonding is a
common alternative to prevent chip-to-chip separation; how-
ever, minimizing the value of DP is indeed essential to avoid
excessive bulging and the ultimate rupture of the Si3N4
membranes.
3.2. NFC optimized for high flow rate
We decided to enhance the robustness of our NFCS by
modifying the channel geometry accounting for all afore-
mentioned technical issues. Our optimized NFC design is
shown in figure 4(a). It circumvents the problem of die-to-die
separation, minimizes window deformation [44], and allows
for high liquid flow rates. As observed from our simulations
in figure 2(b), the internal pressure, Pin, builds up in the inlet
owing to the reduced cross-sectional area (CW SPt ) the liquid
encounters when approaching the window; i.e. SPt is often in
the range of 100 nm to 5 μm.
In order to minimize DP (or Pin), we tuned our fabri-
cation recipes to achieve ECL<30 μm. In addition, it was
crucial to design the shapes of the inlet and outlet channels to
guarantee that DP i–o (DP or Pin) is governed by ECL. In
agreement with equation (1) above [41], our inlet/outlet NFC
channel geometry satisfies L / (b a3)∣ECL / (CW SPt 3);
where L is the length of channel in the NFC, and a and b its
lateral dimensions after ECL with a = b. It should be also
5
A A Petruk et al
pointed out that the implementation of tubing with a very
small inner diameter in the outlet port would also contribute
to increase DP.
Furthermore, the most important feature of our NFC is a
closed inlet channel configuration. Pin builds up in the inlet,
and therefore this is the critical channel region that leads to
chip–chip separation. We solved this issue by ensuring that
our channel geometry in the inlet is such that the resulting
force, rather than separating the dies, brings them together.
An exploded view of our NFC assembly is displayed in
figure 4(a). Two O-rings and a custom-made rubber gasket
provide sealing against the body of the holder tip and the lid.
A stopper made of graphite or teflon protects the top Si-chip
from getting in direct contact with the lid and also confers the
proper gap size to attain about 25% O-ring compression. A
close-up view of the NFC sandwich, stopper and gasket, in
the direction of the liquid flow, is illustrated in figure 4(b).
The arrows correspond to the direction of the resultant force
in the inlet channel. Note that the higher the flow rate, the
greater DP (or Pin), the stronger the force that brings the Si-
chips together is.
Therefore, the liquid flows between the inner holder
surface and the outer surface of bottom Si-die until it
encounters a very small opening, which brings it to the
viewing window via a V-terminated inlet channel. Photo-
graphs of the inner surface of the bottom Si-chip taken right
after dicing are shown in figures 4(c), (d); some solvent
residues and Si3N4 particulates are still visible. The small
opening and the window are discernible in figure 4(d). The
small rectangular freestanding Si3N4 window (ws ≈10 μm) is
in the center of an effective channel with ECL≈30 μm.
Figure 4(e) exhibits the first holder prototype for in-liquid
TEM developed at UeIL. Newer designs will comprise a
modular holder tip with all exchangeable components and
electrical connections for in situ electrochemistry.
3.3. Imaging by in-liquid STEM
We finally demonstrate the application of our NFCS by
imaging gold NPs and residual NPs by STEM (electron
microscope: Hitachi HD2000, acceleration voltage 200 kV).
Figure 5 depicts some representative in-liquid STEM results
obtained at UeIL. In all cases, we started with an empty and
dry NFCS. The main goal in these experiments was to show
that dies do not separate at high flow rate, as a proof of
concept of the proposed NFC design. We were able to obtain
a spatial resolution of about 5 nm in a reproducible manner
while withstanding a flow rate of 2 μl min−1 that guaranteed
>1 kHz for our NFC channel geometry. It should be men-
tioned that in order to attain a decent spatial resolution the
flow had to be stopped. Even a small flow rate of 0.1 μl min−1
was found to be too high for STEM measurements. Our
limited degree of success relied on the fact that all imaged
NPs were stuck onto the window(s). As soon as NPs
detached, they quickly disappeared from the field of view (see
supplementary videos). The NFC configuration implemented
in these studies had 50 nm thick Si3N4 windows, ws ≈17 μm,
ECL≈60 μm, and SPt=500 nm. Figure 5(a) shows the
Nanotechnology 30 (2019) 395703 A A Petruk et al

Figure 4. NFC with closed-inlet channel geometry for extreme liquid flow conditions. (a) Exploded view of NFC assembly in the tip of the
TEM holder. Two small circular holes are visible on the inner surface of the TEM holder tip. These connect via internal tubing that run in the
inside the body of the holder to a syringe pump and an evacuated container. (b) View of the NFC sandwich along the direction of the liquid
flow (perpendicular to the page). The resultant force exerted by Pin in the inlet region is denoted by arrows. (c), (d) Photographs of the inner
surface of the bottom Si-die taken after dicing. A small opening connects the liquid with a V-terminated inlet followed by a very short step or
ECL≈30 μm. The freestanding Si3N4 window with slit shape and ws≈10 μm is in the center of this small step. The blue scale bar is
100 μm. (e) NFC holder designed for Hitachi HD2000 STEM.

formation of gold NPs by electron beam irradiation of mother
liquor. We were capable of imaging only those NPs that were
attached to the window(s). Au NPs were found to form a
quasi-continuous film, which started to detach by 90 s due to
the liquid flow that runs from right to left. The formation of
NPs on the inner surface of the bottom window is also
noticeable.
Figures 5(b), (c) illustrate some images of residual NPs
(unknown composition), which formed by electron beam
irradiation (figure 5(b)) or appeared in solution (figure 5(c))
even after carefully washing the tubing with a diluted solu-
tions of different acids and water. In both cases, we were able
to image NPs that were stuck to the window(s) with (panel
(b)) and without (panel (c)) liquid flow to resolutions that
ranged from 4 to 8 nm. We believe the resolution in the
experiment shown in figures 5(d), (e) is somewhat dictated by
the fact that the NPs do not have well-defined edges. Addi-
tional experiments employing 10 nm gold NPs in the absence
of flow, provided a spatial resolution of about 1 nm for the
same liquid thickness of 500 nm (see supplementary
information). This resolution is closer to the current limit of
our STEM, which is hampered by electromagnetic noise in
room. We think that resolutions of about 1–5 nm obtained by
bright field STEM imaging with 200 kV electrons are rea-
sonable when considering that the employed NFC had 50 nm
thick Si3N4 windows and SPt=500 nm.
4. Conclusions
We introduced an NFC design with improved channel geo-
metry (closed-inlet configuration) that allows for high sample
refreshing rates >1 kHz with minimal window deformation,
and no chip–chip separation. These prerequisites are essential
to perform time-resolved FED experiments in the near future.
A general and obvious observation is that the flow should be
stopped, or samples attached to the window(s) in order to
obtain images at higher resolution by in-liquid STEM. In
addition, we noticed that high-resolution electron microscopy
requires ultraclean conditions. We found residual NPs and

6
Nanotechnology 30 (2019) 395703 A A Petruk et al

Figure 5. (a) Dark field STEM images of gold NPs produced by electron beam irradiation from mother liquor. We could observe the
formation of a film with various nucleation sites (on both windows, see background at 90 and 180 s). The film detaches because of liquid
friction; the liquid flows from right to left; see SI_video_gold. Scale bar 600 nm. (b) Bright field STEM images of NPs formed by electron
beam irradiation from residues in solution. We observed NP growth on the windows followed by NP detachment and immediate
disappearance in the direction of the flow (right to left); see SI_video_NP. Scale bar 200 nm. (c) Bright field STEM images of residual NPs
moving in solution without flow; see SI_video_NP-no-flow. Scale bar 50 nm. Resolution measured by edge contrast was found to be better
than 10 nm. (d), (e) Intensity profiles for resolution measurements (25%–75% of the slope), along the arrow paths shown in panel (b) (30 s)
and panel (c) (5 s), respectively.

beam induced formation of NPs from residues left in the
tubing even after careful cleaning. Therefore, our newest NFC
holder design features all detachable and exchangeable
components. Current efforts at UeIL are focused on adding
electrodes in order to perform in situ electrochemical mea-
surements as well as additional inlet channels for fluid mixing
during in situ TEM measurements. In addition, access to a
300 kV TEM with EELS capability will be necessary to fur-
ther improve image resolution and perform a better char-
acterization of window bulging in the presence of liquid flow;
this requires the fabrication of another holder.
Canada Foundation for Innovation, the Government of
Ontario, and the Interdisciplinary Research Funds Program
of WIN.
Conflicts of interest
The authors declare that there is a US Provisional Patent
Application, USPTO 62/656,418 (filed, 25 April, 2018),
which is related to this work and has been cited as [39].

ORCID iDs
Acknowledgments
Germán Sciaini https://orcid.org/0000-0002-2120-3996
We would like to acknowledge CMC Microsystems, and
Waterloo Commercialization Office for their support. All
nanofabrication processes were carried out at the Quantum
References
NanoFab (QNF) core cleanroom facility of the University of
Waterloo. We would like to thank Dr Stas Dogel and Dr Chad
Ostrander (HHTC) for technical advice and for supporting the [1] Ross F M 2016 Liquid Cell Electron Microscopy (Cambridge :
fabrication of our first NFC holder prototype. GS would like Cambridge University Press)
[2] de Jonge N and Ross F M 2016 Past, present, and future
to acknowledge support from Canada Research Chair pro- electron microscopy of liquid specimens Liquid Cell
gram. Funding for this project was provided by the National Electron Microscopy (Cambridge: Cambridge University
Science and Engineering Research Council of Canada, Press) pp 3–34

7
Nanotechnology 30 (2019) 395703
[3] Double D D 1973 Some studies of the hydration of portland
cement using high voltage (1 MV) electron microscopy
Mater. Sci. Eng. 12 29–34
[4] Sugi H, Akimoto T, Sutoh K, Chaen S, Oishi N and Suzuki S
1997 Dynamic electron microscopy of ATP-induced myosin
head movement in living muscle thick filaments Proc. Natl
Acad. Sci. USA 94 4378–82
[5] Yuk J M et al High-resolution EM of colloidal nanocrystal
growth using graphene liquid cells Science 336 61–4
[6] Dwyer J R and Harb M 2017 Through a window, brightly: a
review of selected nanofabricated thin-film platforms for
spectroscopy, imaging, and detection Appl. Spectrosc. 71
2051–75
[7] Zheng H, Smith R K, Jun Y, Kisielowski C, Dahmen U and
Alivisatos A P 2009 Observation of single colloidal
platinum nanocrystal growth trajectories Science 324
1309–12
[8] Zeng Z, Liang W-I, Liao H-G, Xin H L, Chu Y-H and Zheng H
2014 Visualization of electrode–electrolyte interfaces in
LiPF6/EC/DEC electrolyte for lithium ion batteries via
in situ TEM Nano Lett. 14 1745–50
[9] Williamson M J, Tromp R M, Vereecken P M, Hull R and
Ross F M 2003 Dynamic microscopy of nanoscale cluster
growth at the solid–liquid interface Nat. Mater. 2 532–6
[10] Liao H-G et al 2014 Facet development during platinum
nanocube growth Science 345 916–9
[11] Liao H-G, Cui L, Whitelam S and Zheng H 2012 Real-time
imaging of Pt3Fe nanorod growth in solution Science 336
1011–4
[12] Jungjohann K L, Bliznakov S, Sutter P W, Stach E A and
Sutter E A 2013 In situ liquid cell electron microscopy of the
solution growth of Au–Pd core–shell nanostructures Nano
Lett. 13 2964–70
[13] Alloyeau D et al 2015 Unravelling kinetic and thermodynamic
effects on the growth of gold nanoplates by liquid
transmission electron microscopy Nano Lett. 15 2574–81
[14] White E R, Mecklenburg M, Singer S B, Aloni S and
Regan B C 2011 Imaging nanobubbles in water with
scanning transmission electron microscopy Appl. Phys.
Express 4 055201
[15] Woehl T J and Prozorov T 2015 The mechanisms for
nanoparticle surface diffusion and chain self-assembly
determined from real-time nanoscale kinetics in liquid
J. Phys. Chem. C 119 21261–9
[16] Ye X et al 2016 Single-particle mapping of nonequilibrium
nanocrystal transformations Science 354 874–7
[17] Dukes M J, Jacobs B W, Morgan D G, Hegde H and Kelly D F
2013 Visualizing nanoparticle mobility in liquid at atomic
resolution Chem. Commun. 49 3007–9
[18] Zheng H, Claridge S A, Minor A M, Alivisatos A P and
Dahmen U 2009 Nanocrystal diffusion in a liquid thin film
observed by in situ transmission electron microscopy Nano
Lett. 9 2460–5
[19] Proetto M T et al 2014 Dynamics of soft nanomaterials
captured by transmission electron microscopy in liquid
water J. Am. Chem. Soc. 136 1162–5
[20] Peckys D B, Veith G M, Joy D C and de Jonge N 2009
Nanoscale imaging of whole cells using a liquid enclosure
and a scanning transmission electron microscope PLoS One
4 e8214
[21] Firlar E, Çınar S, Kashyap S, Akinc M and Prozorov T 2015
Direct visualization of the hydration layer on alumina
nanoparticles with the fluid cell stem in situ Sci. Rep. 5
9830
[22] Liu K-L et al 2008 Novel microchip for in situ TEM imaging
of living organisms and bio-reactions in aqueous conditions
Lab Chip 8 1915–21
[23] Riley F L 2000 Silicon nitride and related materials J. Am.
Ceram Soc. 83 245–65
A A Petruk et al
[24] Yang J and Paul O 2002 Fracture properties of LPCVD silicon
nitride thin films from the load-deflection of long
membranes Sensors Actuators A 97–98 520–6
[25] Sciaini G and Miller R J D 2011 Femtosecond electron
diffraction: heralding the era of atomically resolved
dynamics Rep. Prog. Phys. 74 096101
[26] Caroline A 2017 Design and Fabrication of a Nanofluidic Cell
System for High-Resolution Electron Microscopy of in-
Liquid Samples University of Waterloo https://uwspace.
uwaterloo.ca/handle/10012/12146?show=full
[27] de Jonge N, Peckys D B, Kremers G J and Piston D W 2009
Electron microscopy of whole cells in liquid with nanometer
resolution Proc. Natl Acad. Sci. USA 106 2159–64
[28] Grogan J M and Bau H H 2010 The nanoaquarium: a platform
forin situtransmission electron microscopy in liquid media J
Microelectromech. Syst. 19 885–94
[29] Evans J E, Jungjohann K L, Browning N D and Arslan I 2011
Controlled growth of nanoparticles from solution with in situ
liquid transmission electron microscopy Nano Lett. 11 2809–13
[30] Nielsen M H, Lee J R I, Hu Q, Han T Y-J and De Yoreo J J
2013 Structural evolution, formation pathways and energetic
controls during template-directed nucleation of CaCO3
Faraday Discuss. 159 105–21
[31] Chee S W, Pratt S H, Hattar K, Duquette D, Ross F M and Hull R
2014 Studying localized corrosion using liquid cell
transmission electron microscopy Chem. Commun. 51 168–71
[32] Nielsen M H, Aloni S and De Yoreo J J 2014 In situ TEM
imaging of CaCO3 nucleation reveals coexistence of direct
and indirect pathways Science 345 1158–62
[33] Smeets P J M, Cho K R, Kempen R G E,
Sommerdijk N A J M and De Yoreo J J 2015 Calcium
carbonate nucleation driven by ion binding in a biomimetic
matrix revealed by in situ electron microscopy Nat. Mater.
14 394–9
[34] Grogan J M, Schneider N M, Ross F M and Bau H H 2014
Bubble and pattern formation in liquid induced by an
electron beam Nano Lett. 14 359–64
[35] Mueller C, Harb M, Dwyer J R and Miller R J D 2013
Nanofluidic cells with controlled pathlength and liquid flow
for rapid, high-resolution in situ imaging with electrons
J. Phys. Chem. Lett. 4 2339–47
[36] Zečević J, Hermannsdörfer J, Schuh T, de Jong K P and
de Jonge N 2017 Anisotropic shape changes of silica
nanoparticles induced in liquid with scanning transmission
electron microscopy Small 13 1602466
[37] White E R, Mecklenburg M, Shevitski B, Singer S B and
Regan B C 2012 Charged nanoparticle dynamics in water
induced by scanning transmission electron microscopy
Langmuir 28 3695–8
[38] Gardiner D S, Walden F S II and Damiano J Jr 2016 Method
for optimizing fluid flow across a sample within an electron
microscope sample holder US9466459B2
[39] Petruk A A and Sciaini G 2018 Provisional US patent
application: nanofluidic cell systemUSPTO 62/656,418
[40] COMSOL Multiphysics® v.5.3 COMSOL AB Stockholm
https://comsol.com/
[41] White F M 2006 Fluid Mechanics 6th edn (New York:
McGraw-Hill)
[42] Temple-Boyer P, Rossi C, Saint-Etienne E and Scheid E 1998
Residual stress in low pressure chemical vapor deposition
SiNx films deposited from silane and ammonia J. Vac. Sci.
Technol. A 16 2003–7
[43] Merle B and Göken M 2011 Fracture toughness of silicon
nitride thin films of different thicknesses as measured by
bulge tests Acta Mater. 59 1772–9
[44] Tabata O, Kawahata K, Sugiyama S and Igarashi I 1989
Mechanical property measurements of thin films using load-
deflection of composite rectangular membranes Sensors
Actuators 20 135–41