Storing and Reading Information in Mixtures of Fluorescent Molecules

Abstract

The rapidly increasing use of digital technologies requires the rethinking of methods to store data. This work shows that digital data can be stored in mixtures of fluorescent dye molecules, which are deposited on a surface by inkjet printing, where an amide bond tethers the dye molecules to the surface. A microscope equipped with a multichannel fluorescence detector distinguishes individual dyes in the mixture. The presence or absence of these molecules in the mixture encodes binary information (i.e., “0” or “1”). The use of mixtures of molecules, instead of sequence-defined macromolecules, minimizes the time and difficulty of synthesis and eliminates the requirement of sequencing. We have written, stored, and read a total of approximately 400 kilobits (both text and images) with greater than 99% recovery of information, written at an average rate of 128 bits/s (16 bytes/s) and read at a rate of 469 bits/s (58.6 bytes/s).

Full text

Storing and Reading Information in Mixtures of Fluorescent
Molecules
Amit A. Nagarkar, Samuel E. Root, Michael J. Fink, Alexei S. Ten, Brian J. Cafferty,
Douglas S. Richardson, Milan Mrksich, and George M. Whitesides*
Cite This: https://doi.org/10.1021/acscentsci.1c00728
Read Online
ACCESS Metrics & More Article Recommendations *
sıSupporting Information
ABSTRACT: The rapidly increasing use of digital technologies requires
the rethinking of methods to store data. This work shows that digital data
can be stored in mixtures of fluorescent dye molecules, which are
deposited on a surface by inkjet printing, where an amide bond tethers the
dye molecules to the surface. A microscope equipped with a multichannel
fluorescence detector distinguishes individual dyes in the mixture. The
presence or absence of these molecules in the mixture encodes binary
information (i.e., “0”or “1”). The use of mixtures of molecules, instead of
sequence-defined macromolecules, minimizes the time and difficulty of
synthesis and eliminates the requirement of sequencing. We have written,
stored, and read a total of approximately 400 kilobits (both text and
images) with greater than 99% recovery of information, written at an
average rate of 128 bits/s (16 bytes/s) and read at a rate of 469 bits/s (58.6 bytes/s).
■INTRODUCTION
In order to preserve information over long periods of time,
reduce the energy consumption for storage, and prevent
tampering with stored information, new materials and
strategies for storage of information would be useful and
may be required.
1−5
Current devices used to store information
(optical media, magnetic media, and flash memory) have
insufficient operational lifetimes for long-term storage
typically less than two decadesand require substantial energy
to maintain the stored information.
6
Molecules (including, but not limited to DNA) can be used
to store information without power, at high areal density, and
are claimed to be stable for thousands of years or more.
7−14
For these systems to be applied to store information, however,
several problems must be considered including (i) read/write
speeds, (ii) retention of information, (iii) density of
information, and (iv) cost.
15
Here, we demonstrate a write-once-read-many (WORM)
molecular information storage approach using mixtures of
fluorescent dye molecules covalently bound to an epoxy
substrate. An inkjet printer enables writing of information at a
rate of 16 bytes/s, and a multichannel fluorescence detector in
a confocal microscope enables reading at a rate of 58
kilobytes/s. Using this approach, we have written 14 075
bytes of digital information on a 7.2 mm ×7.2 mm surface
(resulting in an aerial information density of 271.5 bytes/mm2)
and read this information over 1000 times without significant
loss (less than 20%) in fluorescent signal intensity. This
approach enables information storage with high density, fast
read/write speeds, and multiple reads of a single set of
molecules without loss of information, all at an acceptable cost.
Devices currently used to store digital information
including optical disks, flash drives, and hard disk drives
have operational lifetimes on the order of decades.
16
An
alternative approach to such technologies is, in principle, to
store information in molecules, as molecule-based storage
systems can have very high theoretical storage densities and
half-lives that can extend to millions of years.
Sequence-defined polymers have been examined for
application in data storage, information processing, and
product validation. Inspired by how nature stores genetic
information, synthetic DNA has become the most popular
molecule to be considered for information storage. While
synthetic DNA provides one of the densest methods of data
storage (∼1018 bytes/mm3),
17
storage of information in long
DNA strands suffers from several significant problems: (i)
DNA sequencing methods (e.g., Next Gen Sequencing
18
) are
slow and, even with massive parallelization, typically require
multiple hours to decode a simple message. This slow rate of
reading makes this technique impractical for many applications
where latency (time to access and read the stored information)
Received: June 17, 2021
Research Article
http://pubs.acs.org/journal/acscii
© XXXX The Authors. Published by
American Chemical Society A
https://doi.org/10.1021/acscentsci.1c00728
ACS Cent. Sci. XXXX, XXX, XXX−XXX
Downloaded via 154.16.30.212 on October 15, 2021 at 18:37:08 (UTC).
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

is important (e.g., data centers); (ii) information systems that
use synthetic DNA typically use polymers that are greater than
100 nucleotides in length, which, due to inefficient monomer
coupling, lead to multiple truncation products that decreases
the information density of the material.
19,20
As an alternative to DNA-based systems, several groups have
examined nonbiological polymers for molecular information
storage. In particular, Lutz and co-workers
11−13
have encoded
binary information into several sequence-defined polymers,
including non-natural polyphosphates,
14
oligo(alkoxyamine
amide)s,
15
and oligo(triazole amide)s
16
and decoded informa-
tion in these polymers by sequencing them with tandem mass
spectrometry. These synthetic polymer systems require
extensive synthesis and purification. For these polymers to
encode kilobytes of data, the polymer chain must be thousands
of units long, but iterative monomer addition suffers from a
decrease in the yield of the polymer with each additional
coupling.
We have recently demonstrated that information can be
stored in the composition of a mixture of oligopeptides, rather
than the sequence of a long polymer with individual units
covalently bonded to form a chain.
21
The use of smaller
fragments, combined with the commercial availability of these
units, eliminates the need for time-consuming and expensive
synthesis. We have used laser-ionization mass spectrometry to
read information stored in molecules on a metal surface. This
method has certain limitations: (i) mass spectrometry is a
destructive approach, and thus information is destroyed during
read-out; (ii) only one location is read at a time, making the
process of read-out slow (20 bits/s) and difficult to parallelize;
(3) there is limited potential to scale down the feature size, as a
decrease in laser spot size leads to an increase in noise.
22
The objective of this work is to demonstrate the storage of
information in a set of optically distinguishable molecules
(rather than oligopeptides distinguishable by molecular weight
using a mass spectrometer). Rather than molecular weight, we
use the difference in the wavelength of fluorescent emission of
commercially available dyestodesignanoptochemical
molecular information storage system. Information is “written”
by inkjet printing of solutions of fluorescent dyes onto a
reactive polymeric substrate. Information is “read”using a
fluorescence microscope equipped with a multichannel
fluorescence detector that can resolve, simultaneously and
independently, any combination of the dyes on the substrate.
This optical read-out technique uses commercially available
technologies and takes advantage of parallelized “reading”. The
system enabled by this combination of molecules is
fundamentally different from other optical storage methods.
The substrate, onto which information is written, is a thin
film of an epoxy polymer, that contains reactive amino groups.
The N-hydroxysuccinimide (NHS) functionalized dyes react
on the substrate to form stable amide bonds. We demonstrate
that these covalently immobilized dyes are stable to more than
1000 reads without significant loss of intensity. In this work,
we used commercially available fluorescent dyes that have been
optimized to reduce the extent of photobleaching.
There are several advantages of our molecular information
storage technique as compared to magnetic tape, which is the
state-of-the-art for long-term storage technique:
27
(i) informa-
tion can be stored, presumably, with lower environmental and
power requirements (in magnetic tape, the binder that secures
the paramagnetic material to the substrate can fail in humid
conditions
28
); (ii) information can be stored less expensively
than with magnetic tape (see Supporting Information, section
S19); (iii) reading of information is parallelizeda single
image file can be used to read the information, unlike
sequential reading in magnetic tape;
29
(iv) information can be
encrypted with novel schemes (see the Registration section).
■RESULTS AND DISCUSSION
Choice of Dye Molecules. We chose seven commercially
available fluorescent dye molecules with different emission
maxima to demonstrate our strategy (Figure 2). The detection
technique, a multichannel fluorescence detector, uses a linear
array of detection channels to resolve multiple emission bands
in parallel and enables spatially resolved information on the
presence or absence of the dye molecules to be obtained in a
single scan across the substrate. In principle, this technique can
be expanded to incorporate more dyes as well (and encode
more information in the same amount of area). The dyes are
dissolved in dimethyl sulfoxide (DMSO), filtered through a
0.45 μm polysulfone syringe filter, and injected into the inkjet
printer cartridge (see Table S1 for concentrations). Figure 2A
lists the dyes used in this study. The optimal concentrations of
the dyes were determined empirically by observing their
fluorescence intensity in a microscope.
“Writing”Information. Inkjet printing is a material
deposition technique that has enabled high-resolution micro-
fabrication with specialized materials and has been demon-
strated to be applicable to areas such as electronics,
30,31
displays,
32,33
drug discovery
34,35
and others.
36
Inkjet printing
has four attractive features: (i) additive operation, where drops
are deposited only where needed; (ii) the ability to use a
variety of inks (aqueous, organic, nanoparticle composites,
biological materials, etc.); (iii) scalability to high throughput
and large substrate area; (iv) lower cost than photo-
lithography-based patterning. We use inkjet printing (other
technologies like aerosol-jet printing
37
and electrohydrody-
namic jet printing
38
provide better printing resolution but are
either too expensive or are not commercially available) to print
Table 1. Comparison of Methods for Archival Data Storage
a
method cost ($/GB) stability write speed
(MB/s) read speed
(MB/s)
magnetic tape
(LTO-7) 0.016
b
10−30 years 4 ×102
c
4×102
c
DNA
23
>530,000
d
up to 2000
years
claimed
e
1×10−3
f
3×10−1
g
SAMDI
21
1
h
not yet
determined 1×10−63×10−6
fluorescent
imaging (this
work)
<0.0001
i
not yet
determined 16 ×10−66×10−2
a
Magnetic tape is the most common technology used to store data for
archival purposes. DNA data storage and self-assembled monolayer-
desorption and ionization (SAMDI) data storage are molecular
information storage strategies that have received interest in the
research community. This work describes storage of information in
mixtures of fluorescent molecules.
b
Total revenue divided by total
data volume of tapes shipped in a year.
24
c
Current generation (LTO-
9).
d
Reports vary ($530,000−$31,250,000 per GB written). DNA
sequencing also incurs cost.
e
Estimated lifetime of DNA encapsulated
in silica.
17
f
Overall throughput is estimated to be on the order of kB/
s.
25
Specific values for writing rates are not reported.
g
Using a single
state-of-the-art sequencing device.
26
h
Self-assembled monolayers for
matrix-assisted laser desorption/ionization;
21
i
See Supporting In-
formation. section S19 for detailed calculations.
ACS Central Science http://pubs.acs.org/journal/acscii Research Article
https://doi.org/10.1021/acscentsci.1c00728
ACS Cent. Sci. XXXX, XXX, XXX−XXX
B

1 pL droplets with a 30-μm center-to-center distance between
adjacent spots on the substrate (Supporting Information,
Figure S4). To demonstrate storage of information at high
density, we wrote the first section of one of the most seminal
research papers in scientific history: “Experimental researches
in electricity”by Michael Faraday, Phil. Trans. R. Soc. Lond.
1832,122, 125−162 (Supporting Information, section S18).
This text contains 14 075 characters (i.e., 14 075 bytes of
information when converted to ASCII).
“Reading”Information. Fluorescence imaging is a
powerful tool for high-resolution characterization of biological
samples and materials. The availability of a variety of
fluorescent dyes enables unprecedented control in the labeling
of specific sites on the sample. Recently, the analysis of spectral
data sets and the separation of signals by spectral imaging,
combined with linear unmixing, have overcome problems of
spectral overlap for fluorescent dyes, and is used widely in
biological systems.
39
We used a Zeiss LSM 800 fluorescence microscope, which
has one of the most versatile implementations for spectral
imaging. In this technique, fluorescence emission passes
through a pinhole and is separated by wavelength by a
diffraction grating (Supporting Information, Figure S5). The
spectrally resolved light is then projected onto a linear array of
34 detection channels in a photomultiplier detector. The
wavelength of emitted light is determined by the position of
the channel receiving the photons. This system allows very
precise determination of the intensity of peaks separated by
only a few nanometers and thus the concentration of the dyes
responsible. The presence/absence of a specificfluorescent dye
molecule at a specific location on the substrate can thus be
determined.
Choice of Substrate. Long-term storage of information
requires the formation of thermodynamically stable bonds with
very long half-lives. An amide bond is one of the most
thermodynamically stable bonds available to organic chem-
ists.
40
In our strategy, we used N-hydroxysuccinimide-
functionalized dye molecules, which spontaneously react with
amino groups to form amide bonds. We synthesized a cross-
linked epoxy polymer with an excess of the amine curing agent.
This substrate contains reactive secondary amino groups
(Supporting Information, Figure S2).
41
The epoxy polymer is
processed by hot-pressing a mixture of bisphenol A diglycidyl
ether and triethylene tetramine at 120 °C between a glass
coverslip and a flat PDMS surface (see Experimental Section).
We control the pressure (70 psi) to obtain 50-μm thick films
(Supporting Information, Figures S2 and S3). It is important
to have a flat surface for the substrate because irregularities in
thickness lead to blurring of the image and incomplete focusing
in the microscope on reading the information (Supporting
Information Figure S16 for an example).
Encoding Scheme for “Writing”of Binary Informa-
tion. A binary representation of ASCII characters consists of
eight bits where each bit is either “0”or “1”.
42
In our encoding
scheme, the binary representation of each ASCII character in
the bit string is assigned a position (positions 1−8, Figure 3).
This position is assigned to a fluorescent dye molecule (here,
we assign dyes to positions in the order of increasing emission
maxima). The bit strings for the positions are then used to
generate a printable pattern. Here, we generate a square
pattern out of the bit strings, but, in principle, any pattern
geometry is possible. These square patterns are then
sequentially printed on the substrate using an inkjet printer.
“0”indicates absence of a dye molecule, and “1”indicates the
presence of a dye molecule. For printable ASCII characters, the
first binary digit is always “0”, and hence, the first square
pattern is always a blank pattern. Thus, we require only seven
dye molecules for data storage of printable eight-bit ASCII
characters.
Registration. Figure 3B shows a schematic representation
of the fact that registration of the printed grids of fluorescent
molecules is not required. The fluorescent molecules, when
deposited onto the substrate, lie on a grid where the presence
or absence of the molecule at the intersection of the gridlines
determines binary information (i.e., “0”or “1”). When these
grids are sequentially printed onto the substrate, any offset
between grids of different fluorescent molecules does not make
adifference to the output obtained on reading through a
multichannel-fluorescence detector. To help in determining
the position of the grid, we place three dots that serve as
“calibration spots”as shown in Figure 3B.
We used the Fujifilm Dimatix DMP 2831 inkjet printer to
deposit the dye molecules onto the substrate. As this inkjet
printer can accommodate only one cartridge at a time, we
manually changed the cartridges (each containing one
fluorescent dye solution) to print the computer-generated
images. We needed 7 manual cartridge changes, and it took
116 s on an average to write each pattern for “Experimental
researches in electricity”at 30 μm center-to-center spot
distance on a 7.2 mm ×7.2 mm substrate area. This time and
area translate into a writing speed of 16 bytes/s.
The substrate with the written information was placed in a
Zeiss LSM 880 fluorescence microscope in an inverted
Figure 1. A schematic diagram of the “writing”and “reading”process.
ASCII information is converted to a binary bit string which is then
encoded into printable patterns and printed with an inkjet printer.
The presence or absence of dye molecules at a location represents a
byte of data. The information is written on an epoxy substrate which
contains free amino groups. Printing of the dyes leads to an amide
bond formation between the substrate and the dye and leads to
covalent immobilization of the dye onto the substrate at a specific
location. Imaging of the printed substrate using a multichannel
fluorescence detector represents the “reading”of the written
information. The multichannel fluorescence detector can, simulta-
neously and independently, detect the presence or absence of the dye
molecules at a specific location. One very important feature of our
approach is that the registration of the dyes with respect to each other
is not important for decoding the stored information.
ACS Central Science http://pubs.acs.org/journal/acscii Research Article
https://doi.org/10.1021/acscentsci.1c00728
ACS Cent. Sci. XXXX, XXX, XXX−XXX
C

configuration. Four lasers (405 nm, 488 nm, 561 nm, 633 nm)
were chosen to excite all the dyes simultaneously in the visible
spectrum region (410−695 nm). Using the in-built spectral
imaging function, we could resolve all the patterns with very
good spatial resolution for each dye. Figure 4B show cropped
regions of the unmixed images for all seven dye molecules. It
took approximately 240 s to record the image, giving an
effective reading speed of 58.64 kilobytes/s (469 kilobits/s).
Here we use an inkjet printer and print at the highest
resolution, where it is not possible to specify whether the grids
are offset or perfectly overlap. The multichannel fluorescence
detector is required to show the presence/absence of a dye in
the same location as other dyes if they perfectly overlap.
Decoding the Information. It is straightforward to use
image analysis to decode the stored information. The
individual patterns are read using a simple Python program
script using the OpenCV computer vision library.
43
We
obtained good accuracy (99.64%) of the recovered information
(measured as the number of bits read correctly as a percentage
of the total number of bits). This accuracy can be improved
with more sophisticated image analysis techniques and error
correction codes.
44
The most common reason for inaccuracies
during reading were dust particles adhering to the substrate
surface.
Stability of the Information. Photobleaching is the
attenuation of fluorescence intensity of a fluorophore molecule,
primarily due to the cleavage of covalent bonds in the molecule
on reaction with oxygen. In our experiments, photobleaching
did not significantly affect our recorded information. As
compared to traditional biological labeling experiments, we use
a high concentration of the fluorescent dye (micromolar
quantities). Two benefits of using high concentration of dyes
are (i) low laser power is required to excite the fluorescent
dyes at a location, (ii) lower laser power also decreases the rate
of photobleaching. In our experiments, a 2 mm ×2mm
portion of the information was continuously read 1000 times
in air without significant loss in intensity (Figure 5). After 1000
reading cycles, dye 425 showed the largest reduction in
intensity (∼21%), while all other dyes showed a <15% change
in intensity.
Storage of Digital Images. Our strategy of storage of
information for ASCII data can be applied to store non-ASCII
data as well. As shown in Figure 6, we converted a 3 kilobyte
JPEG image of Michael Faraday into a bit string, encoded the
bit string to print in seven fluorescent dyes, and inkjet printed
the molecules onto the substrate. In this case, as the data are
already in a compressed format (JPEG), the quality of
recovered data is much more sensitive to errors than when it
is in a loss-less image encoding format. An example of the
image with 0.4% printing errors (0.4% bits read wrong as
compared to the input bit string) is given in the Supporting
Information (Figure S17).
Our technology for storage of information in mixtures of
fluorescent molecules can be expanded with the use of other
fluorophores with narrow emission bandwidths (e.g., quantum
dots
45
or J-aggregates
46
). An expanded palette of fluorophores
will allow for the use of more fluorophores per location and
could also allow simpler, band-pass filter-based reading,
eliminating the requirement for an expensive multichannel
fluorescence detector. More sophisticated drop-on-demand
technologies (electrohydrodynamic inkjet printer commercial-
ized by SIJ corporation, Japan, dip-pen lithography,
47
etc.) can
print at much higher resolutions (sub-1 μm spot−spot
distance). At 1 μm spot-to-spot distance with eight fluorescent
Figure 2. Optically distinguishable fluorescent dyes. (A) Structures of the fluorescent dyes used in this study along with their emission spectra in
dimethyl sulfoxide. (B) Reaction scheme for the covalent immobilization of the dye molecule on the substrate. Amino groups in the substrate react
with the N-hydroxysuccinimide derivatives of the fluorescent dye to link the fluorescent dye to the substrate with an amide bond.
ACS Central Science http://pubs.acs.org/journal/acscii Research Article
https://doi.org/10.1021/acscentsci.1c00728
ACS Cent. Sci. XXXX, XXX, XXX−XXX
D

dyes, the areal storage density will be 5 Gbits/in2, which is
comparable to the latest generation of magnetic tape (LTO-8,
areal density: 8 Gbit/in2). Another area for improvement is the
use of error correction codes to decrease error rates (e.g., Reed
Solomon error correction codes
44
have been extensively used
to decrease error rates in optical media like compact discs, and
Blu-ray discs).
■CONCLUSION
In conclusion, we report a fundamentally new molecular data
storage technology that leverages the optical characteristics of
conjugated molecules. A multichannel fluorescence detector
enables the simultaneous and independent detection of the
presence or absence of a molecule in a mixture on a surface.
The “writing”process uses inkjet printing wherein molecules
that are deposited onto the surface form an amide bond to link
the dye molecules to the substrate. An important characteristic
of this information storage method is that registration of the
individual molecules is not required. This characteristic is, to
our best knowledge, unique; it differentiated this method from
existing optical data storage technologies.
We also show that multiple (>1000) readouts of such optical
molecular information are possible without significant loss of
information through bleaching and other mechanisms. This is
also unique as compared to other molecular information
storage systems which involve destructive reading (e.g.,
sequencing of DNA
17
or laser-ablation of oligopeptides
21
).
We have also demonstrated the fastest reading speed of any of
the molecular information storage methods (0.469 Mbits/s).
Access to newer drop-on-demand technologies like electro-
hydrodynamic inkjet printing would enable commercially
competitive areal information density.
This optical molecular information storage technology
presents solutions to important problems that are faced by
emerging molecular information storage technologies: energy
used for storage, cost, and ability to resist corruption.
■EXPERIMENTAL SECTION
Safety. Epoxy resins are known skin sensitizers and should
be handled carefully with all safety precautions and personal
protective equipment in a well-ventilated fume hood. Caution
must be taken while handling the reactive fluorescent dyes as
their safety hazards are not fully known.
Materials. AlexaFluor dyes were purchased from Thermo
Fisher and used without further purification. Atto 425 dye, dry
dimethyl sulfoxide (DMSO), bisphenol A diglycidyl ether, and
triethylenetetramine were purchased from Sigma-Aldrich and
used without further purification.
Fabrication of the Epoxy Substrate. Bisphenol A
diglycidyl ether (2.4 g, 7 mmol) was mixed with 0.6 g of
triethylene tetramine (4.2 mmol, 3 equiv). This solution was
vigorously stirred for 2 min and degassed under a vacuum (80
mbar) for 5 min. This solution (2.6 mL) was poured onto a
glass slide and placed inside a heat-press. A PDMS (Sylgard
184) block (10 cm ×10 cm ×0.5 cm) was placed on top of
this solution. The PDMS slab plays two roles: (i) it ensures
that the top surface is flat, (ii) it does not stick to the epoxy
film, and hence it is easy to remove after the epoxy polymer has
cured. The polymer was cured under 20 psi pressure at 120 °C
for 30 min. The PDMS layer was manually removed, and the
Figure 3. Encoding and registration. (A) Encoding scheme for storage of data for ASCII characters. The algorithm converts the input ASCII string
into binary bit strings. Each specific position in the binary data is combined to generate a separate bit string, which corresponds to a specific
fluorescent molecule. These eight-bit strings are then converted into a pattern (here, a square pattern, but, in principle, any shape of an array of
spots can be used) and printed sequentially onto the substrate with an inkjet printer. (B) A schematic representation demonstrating that the
registration of different colors of the printed grids is not required. The grids, when printed onto the substrate, can either be perfectly registered or
be printed with an offset. In both cases, as information is read using fluorescence emission at predetermined wavelengths, the patterns can be read
independently (1 = presence of the dye and 0 = absence of the dye). Independent read-out from each channel of the fluorescent detector facilitates
this “non-registered”information storage.
ACS Central Science http://pubs.acs.org/journal/acscii Research Article
https://doi.org/10.1021/acscentsci.1c00728
ACS Cent. Sci. XXXX, XXX, XXX−XXX
E

epoxy film on the glass substrate was cooled down to room
temperature. This film was then washed with n-hexane three
times to remove any potential residue left by the PDMS
polymer.
Instrumentation. Inkjet Printing. Writing was carried out
with a Fujifilm Dimatix DMP 2831 printer with a 1 pL printing
volume cartridge. The printing parameters are firing voltage of
the active nozzle: 16 V; firing voltage of inactive nozzles: 12 V,
printing height: 0.5 mm. Cartridges containing the fluorescent
dyes were manually changed for each dye.
Atomic Force Microscopy. A sample consisting of an epoxy
film on a glass slide was first sonicated in isopropyl alcohol for
10 min and then dried under a nitrogen stream. Atomic force
microscope (AFM) images were obtained using an Asylum
Research Cypher AFM in tapping mode with a 300 kHz
cantilever. Supporting Information, Figure S2 provides the
data.
Profilometry. A sample consisting of an epoxy film on a
microscope glass slide (VWR, 1 mm thickness) was sliced with
a razor blade to introduce trenches into the film, sonicated in
isopropyl alcohol for 10 min, and then dried under nitrogen.
Profilometry was performed using a Bruker DektakXT
profilometer equipped with a 5-μm radius diamond tip and
with 3 mg of applied force.
Supporting Information, Figure S3 provides the data.
Microscopy. Reading was carried out using a Zeiss LSM 880
confocal microscope with an in-built 34 channel photo-
multiplier detector. Four lasers were used to excite all the dyes:
Figure 4. Reading of information. (A) Fluorescence microscope image of the first section of Faraday’s“Experimental researches in electricity”on a
50 μm epoxy polymer film written using the encoding scheme shown in Figure 3. The image was recorded with excitation using four lasers
simultaneously (405 nm, 488 nm, 561 nm, 633 nm). (B) Zoomed-in image of the printed droplets on the epoxy substrate. (C) Linear unmixing of
the fluorescent microscope image leads to independent deconvolution of each dye at a location. The panel shows individual grids of fluorescent dye
molecules 425, 488, 514, 555, 568, 594, and 647 obtained by spectral unmixing of the original image using the Zeiss Zen Black software.
Figure 5. A subset of the printed area was continuously read 1000 times in the fluorescence microscope. (A) Image of the printed region on the
first read. (B) Image of the same region after 1000 reads. All the patterns of the dyes were easily readable after 1000 cycles of reading. (C) Table
showing the percentage of the initial fluorescence intensity remaining after reading the data 1000 times in air.
ACS Central Science http://pubs.acs.org/journal/acscii Research Article
https://doi.org/10.1021/acscentsci.1c00728
ACS Cent. Sci. XXXX, XXX, XXX−XXX
F

405 nm, 488 nm, 561 nm, 633 nm. The in-built multichannel
fluorescence detector was calibrated with individual dyes
printed on the epoxy substrate.
■ASSOCIATED CONTENT
*
sıSupporting Information
The Supporting Information is available free of charge at
https://pubs.acs.org/doi/10.1021/acscentsci.1c00728.
Absorption and emission spectra of fluorescent dyes, list
of concentrations of the fluorescent dyes used, character-
ization of the substrate, spectrally unmixed images of
each dye, computer-generated printing patterns, images
of printed droplets, image of problems while focusing in
the microscope, stored and decoded image of Michael
Faraday containing errors, encoded text from Faraday’s
“Experimental Researches in Electricity”, estimation of
the lower bound of cost per GB, typical inkjetting
waveform, and an estimation of lifetime of the
fluorescent dyes (1010 years by extrapolation using the
Arrhenius equation) (PDF)
■AUTHOR INFORMATION
Corresponding Author
George M. Whitesides −Department of Chemistry and
Chemical Biology, Harvard University, Cambridge,
Massachusetts 02138, United States; orcid.org/0000-
0001-9451-2442; Email: gwhitesides@
gmwgroup.harvard.edu
Authors
Amit A. Nagarkar −Department of Chemistry and Chemical
Biology, Harvard University, Cambridge, Massachusetts
02138, United States
Samuel E. Root −Department of Chemistry and Chemical
Biology, Harvard University, Cambridge, Massachusetts
02138, United States
Michael J. Fink −Department of Chemistry and Chemical
Biology, Harvard University, Cambridge, Massachusetts
02138, United States; orcid.org/0000-0003-0023-8767
Alexei S. Ten −Department of Chemistry and Chemical
Biology, Harvard University, Cambridge, Massachusetts
02138, United States; orcid.org/0000-0003-0040-9132
Brian J. Cafferty −Department of Chemistry and Chemical
Biology, Harvard University, Cambridge, Massachusetts
02138, United States
Douglas S. Richardson −Harvard Center for Biological
Imaging, Cambridge, Massachusetts 02138, United States
Milan Mrksich −Department of Chemistry and Department
of Biomedical Engineering, Northwestern University,
Evanston, Illinois 60208, United States; orcid.org/0000-
0002-4964-796X
Complete contact information is available at:
https://pubs.acs.org/10.1021/acscentsci.1c00728
Author Contributions
G.M.W. conceived the idea of molecular information storage in
mixtures of molecules. A.A.N., B.J.C., S.E.R. and G.M.W.
postulated information storage in mixtures of fluorescent
molecules. A.A.N., S.E.R., and D.R. conducted the experiments
and performed the analysis. A.S.T., M.J.F., and M.M. provided
valuable input for improvement of the manuscript. A.A.N.,
S.E.R., M.J.F., and G.M.W. wrote the manuscript with inputs
from all the authors.
Notes
Theauthorsdeclarethefollowingcompetingfinancial
interest(s): A.A.N., A.S.T., and M.J.F. acknowledge an equity
interest in Datacule Inc. G.M.W. acknowledges an equity
interest and a board position in Datacule Inc.
■ACKNOWLEDGMENTS
This work was supported by Defence Advanced Research
Projects Agency (DARPA) under Award No. W911NF-18-2-
0030.
■REFERENCES
(1) Lloyd, S. Ultimate physical limits to computation. Nature 2000,
406, 1047−1054.
(2) Shulaker, M. M.; et al. Three-dimensional integration of
nanotechnologies for computing and data storage on a single chip.
Nature 2017,547,74−78.
(3) Pansare, A. V.; et al. In Situ Nanoparticle Embedding for
Authentication of Epoxy Composites. Adv. Mater. 2018,30, 1801523.
(4) Salahuddin, S.; Ni, K.; Datta, S. The era of hyper-scaling in
electronics. Nature Electronics 2018,1, 442−450.
(5) Baliga, J.; Ayre, R. W.; Hinton, K.; Tucker, R. S. Green Cloud
Computing: Balancing Energy in Processing, Storage, and Transport.
Proc. IEEE 2011,99, 149−167.
(6) Brandner, R.; Pordesch, U.; Wallace, C. Long-Term Archive
Service Requirements; The IETF Trust: Internet Requests for
Comments, 2007.
(7) Clelland, C. T.; Risca, V.; Bancroft, C. Hiding messages in DNA
microdots. Nature 1999,399, 533−534.
Figure 6. A JPEG image of Michael Faraday (3.95 KB) was converted into patterns for the seven fluorescent dyes used in this study. This
information was printed onto a 4.2 mm ×4.2 mm epoxy substrate. An example of a decoded image with 0.4% printing errors is shown in the
Supporting Information, Figure S17. The image of Michael Faraday has been reproduced with permission from Getty Images.
ACS Central Science http://pubs.acs.org/journal/acscii Research Article
https://doi.org/10.1021/acscentsci.1c00728
ACS Cent. Sci. XXXX, XXX, XXX−XXX
G

(8) Green, J. E.; et al. A 160-kilobit molecular electronic memory
patterned at 1011 bits per square centimetre. Nature 2007,445, 414−
417.
(9) Colquhoun, H.; Lutz, J.-F. Information-containing macro-
molecules. Nat. Chem. 2014,6, 455−456.
(10) Zhirnov, V.; Zadegan, R. M.; Sandhu, G. S.; Church, G. M.;
Hughes, W. L. Nucleic acid memory. Nat. Mater. 2016,15, 366−370.
(11) Al Ouahabi, A.; Charles, L.; Lutz, J.-F. Synthesis of non-natural
sequence-encoded polymers using phosphoramidite chemistry. J. Am.
Chem. Soc. 2015,137 (16), 5629−5635.
(12) Charles, L.; Laure, C.; Lutz, J.-F.; Roy, R. K. MS/MS
sequencing of digitally encoded poly (alkoxyamine amide)s. Macro-
molecules 2015,48 (13), 4319−4328.
(13) Amalian, J.-A.; Trinh, T. T.; Lutz, J.-F.; Charles, L. MS/MS
digital readout: analysis of binary information encoded in the
monomer sequences of poly (triazole amide)s. Anal. Chem. 2016,
88 (7), 3715−3722.
(14) Goodwin, C. A.; Ortu, F.; Reta, D.; Chilton, N. F.; Mills, D. P.
Molecular magnetic hysteresis at 60 K in dysprosocenium. Nature
2017,548, 439−442.
(15) Bhat, W. A. Bridging data-capacity gap in big data storage.
Future Generation Computer Systems 2018,87, 538−548.
(16) Gu, M.; Li, X.; Cao, Y. Optical storage arrays: a perspective for
future big data storage. Light: Sci. Appl. 2014,3, e177.
(17) Grass, R. N.; Heckel, R.; Puddu, M.; Paunescu, D.; Stark, W. J.
Robust Chemical Preservation of Digital Information on DNA in
Silica with Error-Correcting Codes. Angew. Chem., Int. Ed. 2015,54,
2552−2555.
(18) Goodwin, S.; McPherson, J. D.; McCombie, W. R. Coming of
age: ten years of next-generation sequencing technologies. Nat. Rev.
Genet. 2016,17, 333−351.
(19) Kosuri, S.; Church, G. M. Large-scale de novo DNA synthesis:
technologies and applications. Nat. Methods 2014,11, 499−507.
(20) Organick, L.; et al. Random access in large-scale DNA data
storage. Nat. Biotechnol. 2018,36, 242−248.
(21) Cafferty, B. J.; Ten, A. S.; Fink, M. J.; Morey, S.; Preston, D. J.;
Mrksich, M.; Whitesides, G. M. Storage of Information Using Small
Organic Molecules. ACS Cent. Sci. 2019,5(5), 911−916.
(22) Feenstra, A. D.; Dueñas, M. E.; Lee, Y. J. Five Micron High
Resolution MALDI Mass Spectrometry Imaging with Simple,
Interchangeable, Multi-Resolution Optical System. J. Am. Soc. Mass
Spectrom. 2017,28, 434−442.
(23) Catalogue Press Release, https://techthelead.com/start-up-
stores-all-16g-of-wikipedia-onto-dna-strand/.
(24) Fontana, R. E.; Decad, G. M. Moore’s law realities for recording
systems and memory storage components: HDD, tape, NAND, and
optical. AIP Adv. 2018,8, 056506.
(25) Ceze, L.; Nivala, J.; Strauss, K. Molecular digital data storage
using DNA. Nat. Rev. Genet. 2019,20, 456−466.
(26) Meiser, L. C.; Koch, J.; Antkowiak, P. L.; Stark, W. J.; Heckel,
R.; Grass, R. N. DNA synthesis for true random number generation.
Nat. Commun. 2020,11, 5869.
(27) Lantz, M. Why the Future of Data Storage Is (Still) Magnetic
Tape. IEEE Spectrum: Technology Engineering and Science News, Aug.
2018.
(28) Bertram, H.; Cuddihy, E. Kinetics of the humid aging of
magnetic recording tape. IEEE Trans. Magn. 1982,18, 993−999.
(29) Sandsta, O.; Midtstraum, R. Analysis of retrieval of multimedia
data stored on magnetic tape. Proceedings International Workshop on
Multi-Media Database Management Systems; Cat. No. 98TB100249;
Dayton, OH, USA, 1998; pp 54−63.
(30) Sirringhaus, H.; et al. High-resolution inkjet printing of all-
polymer transistor circuits. Science 2000,290, 2123−2126.
(31) Shimoda, T.; et al. Solution-processed silicon films and
transistors. Nature 2006,440, 783−786.
(32) Shimoda, T.; Morii, K.; Seki, S.; Kiguchi, H. Inkjet printing of
light-emitting polymer displays. MRS Bull. 2003,28, 821−827.
(33) Chang, S. C.; et al. Multicolor organic light-emitting diodes
processed by hybrid inkjet printing. Adv. Mater. 1999,11, 734−737.
(34) Lemmo, A. V.; Rose, D. J.; Tisone, T. C. Inkjet dispensing
technology: Application in drug discovery. Curr. Opin. Biotechnol.
1998,9, 615−617.
(35) Heller, M. J. DNA microarray technology: Devices, systems,
and applications. Annu. Rev. Biomed. Eng. 2002,4, 129−153.
(36) Hiller, J.; Mendelsohn, J. D.; Rubner, M. F. Reversibly erasable
nanoporous anti-reflection coatings from polyelectrolyte multilayers.
Nat. Mater. 2002,1,59−63.
(37) Wilkinson, N. J.; Smith, M. A.; Kay, R. W.; Harris, R. A. A
review of aerosol jet printinga non-traditional hybrid process for
micro-manufacturing. International Journal of Advanced Manufacturing
Technology 2019,105, 4599−4619.
(38) Park, J.-U.; et al. High-resolution electrohydrodynamic jet
printing. Nat. Mater. 2007,6, 782−789.
(39) Valm, A. M.; Oldenbourg, R.; Borisy, G. G. Multiplexed
Spectral Imaging of 120 Different Fluorescent Labels. PLoS One 2016,
11, e0158495.
(40) Martin, R. B. Free energies and equilibria of peptide bond
hydrolysis and formation. Biopolymers 1998,45 (5), 351−353.
(41) Pansare, A. V.; et al. Shape-Coding”: Morphology-Based
Information System for Polymers and Composites. ACS Appl. Mater.
Interfaces 2020,12, 27555−27561.
(42) Hieronymus, J. L. ASCII phonetic symbols for the world’s
languages Worldbet. J. Int. Phonetic Assoc. 1993,23.
(43) Pulli, K.; Baksheev, A.; Kornyakov, K.; Eruhimov, V. Realtime
Computer Vision with OpenCV. Queue 2012,10,40−56.
(44) Reed, I. S.; Solomon, G. Polynomial Codes over Certain Finite
Fields. J. Soc. Ind. Appl. Math. 1960,8(2), 300−304.
(45) Owen, J.; Brus, L. Chemical Synthesis and Luminescence
Applications of Colloidal Semiconductor Quantum Dots. J. Am.
Chem. Soc. 2017,139 (32), 10939−10943.
(46) Würthner, F.; Kaiser, T. E.; Saha-Möller, C. R. J-Aggregates:
From Serendipitous Discovery to Supramolecular Engineering of
Functional Dye Materials. Angew. Chem., Int. Ed. 2011,50, 3376−
3410.
(47) Liu, G.; Petrosko, S. H.; Zheng, Z.; Mirkin, C. A. Evolution of
Dip-Pen Nanolithography (DPN): From Molecular Patterning to
Materials Discovery. Chem. Rev. 2020,120, 6009−6047.
ACS Central Science http://pubs.acs.org/journal/acscii Research Article
https://doi.org/10.1021/acscentsci.1c00728
ACS Cent. Sci. XXXX, XXX, XXX−XXX
H