Single Molecule and Single Cell Sensing with Nanoelectromechanical Systems (NEMS)
Sep 5, 2023
"Single Molecule and Single Cell Sensing with Nanoelectromechanical Systems (NEMS)"
Michael Roukes, Professor of Physics, Applied Physics, and Bioengineering, and Co-Director of the Kavli Nanoscience Institute, California Institute of Technology
CNST Nanotechnology Workshop 2013
And we're going to start the second half of this morning
session with another keynote speaker.
This is Michael Rukus, who is a professor of physics,
applied physics and bioengineering,
and was the founding director of Caltech's Kavli.
Is that how you say it?
Nanoscience Institute.
He's going to speak to us today about single molecule
and single cell sensing with nano electromechanical systems.
You're scheduled for 45 minutes, but it's interesting enough.
You can talk as much as you like.
It's impossible to do that.
Yeah, can you dim the lights a little bit?
Perfect.
Thanks a lot.
It's a pleasure to be here and to see some old friends.
I thought I would tell you a little bit about some work
that we've been doing for, it's like going on 14 years,
I think, now.
Trying to take promising nano devices
and turn them into systems that can actually
be useful for biological measurements.
And the 14 years maybe, in itself,
is a sobering tale about what's involved
in trying to go from some concept of devices
and initial papers where you say,
this has great promise for doing interesting things,
to actually doing interesting things.
And so we're starting just to get
to the phase of doing interesting things,
as I'll try to convince you.
So the story starts with an ultra high frequency
nano electromechanical systems resonator.
There's a picture of it here.
The thing that's down at the bottom, labeled C,
is a doubly clamped beam that's suspended
above a silicon substrate.
And if you make measurements of the resonance of this beam
electrically, in this particular case,
this was read out by magnetomotive transduction.
I can go into details and question period
if you're interested in that kind of stuff.
The resonant frequency for this little device
is just below 429 megahertz.
You can see the amplitude and the phase of the resonance.
So think of this as like a little guitar string that
vibrates at ultra high frequencies.
In order to get good measurements,
we had to devise, over the course of those years
I told you about, ways of suppressing background.
Because often when you try to first measure nanodevices,
background is all you measure.
So after a bunch of time, we got pretty good at it.
And we could look at the resonant frequency
of this device.
That's what's shown here as a function of time,
using a phase lock loop.
So we could read out, have it continuously vibrating,
continuously driven in phase lock,
and look at frequency fluctuations.
And here it shows you the frequency fluctuations
of this thing on this time scale,
are about a part in 10 to the 7, which we worked hard to get
and we're very proud of.
But any frequency control people who
are used to big silicon devices would
think this is totally pathetic, because good frequency control
is a part in 10 to the 14 or so.
Nonetheless, these are incredibly sensitive devices,
as I'll show you.
And this is good enough to get atomic scale mass measurements.
If there's time, I put in a couple of slides at the end
to come back to this question about,
why are nano things so noisy?
So we'll see.
Basically because they're small, if I don't get to it.
But I'll tell you more about it later.
So we decided in 1998, 1999, that one
of the interesting things to do with these little
nanomechanical resonators was to stick atoms to them.
And we did some initial back of the envelope estimates
and we convinced ourselves that, in fact, when
you stick atoms or molecules on there,
you should be able to sense them.
And I won't go through all the history
of building up to the sensitivity that we have today.
But the bottom line is that in around 2006, 2007,
we got our first data where we could actually
see individual molecular adsorption
events in frequency shifts of these device.
So the black is in the absence of a flux of molecules.
And when we turn on a flux of molecules of bovine serum
albumin, which is 66 kilodaltons, or roughly
66,000 AMU, we see every time a molecule accretes
onto the device, and it sticks because we
keep the device cold, just like frost on a window
paint, physisorption.
Every time a molecule accretes on the device,
we see a frequency shift downward
because the resonant frequency, from freshman physics,
goes something like the square root
of the stiffness over the mass.
And if you increase the mass, then the frequency
has to go down.
These are the guys that did it.
The principle of this mass sensing
is then fairly straightforward.
We have a resonant response.
We have, for this little sort of guitar string
that's oscillating at high megahertz range.
Quack.
Sorry about that.
When my wife rings me, there's a beautiful harp playing,
and everything else is a quacking duck.
When the molecules accrete onto the device, as I said, the frequency has to go downward.
And we can think of this sort of formally in terms of an equation where the output is
the frequency shift.
The input is this little change of mass, the accretion from one atomic species or molecular
species onto the device.
And it's proportional in a certain regime to something we think of like a gain or a
responsivity, more formally, this added mass times this mass responsivity gives us a frequency
shift.
And here you can see in real time for these data, background fluctuations, nothing's accreted
yet.
And when the first device comes in, this whole thing rescales and you can see there's the
first atomic absorption event.
Now how good can this actually be?
How far can we push it?
Well, you know, back in early 1999, 2000s, trying to look ahead and see what the date
was of this paper.
I forgot when we published it.
But basically we looked out in the literature and asked what are the ultimate limits, the
ultimate physical limits for doing these kinds of mass measurements.
And we didn't find any.
So we set off to actually do an analysis ourselves.
And essentially developed this formalism and an understanding that the minimum resolvable
mass is inversely proportional to this mass responsivity times noise, frequency fluctuation
noise in the problem.
Because the frequency instability masks our ability to see those jumps downward as a function
of time when atomic species accrete.
And so this paper explores a number of the fundamental limits to mass sensing.
And I'll come back to this at the end of the talk again if there's time.
But we convinced ourselves that it should be possible to resolve basically the accretion
of protons onto the device.
So 1AMU mass sensitivity looked quite feasible to us in the early 2000s.
And so given that, this looks like a really interesting prospect for turning this into
a new type of mass spectrometry.
So how do you minimize the resolvable mass?
Well, you want to make this term in here, the mass responsivity, as big as possible.
So the inverse basically gives us something that's small here, meaning the mass resolution
goes down as far as possible.
And the good news is that for nanoscale, for all types of devices, the mass responsivity
scales basically inversely proportional to, if you scale all dimensions down simultaneously
to a small device, scales inversely proportional to 1 over length to the 4th power.
So what this means is even from going to micro to nano, this L to the 4th will give you 12
orders of magnitude.
The three orders of magnitude in size, and going from MEMS to NEMS makes all the difference.
And so, you know, our early experiments actually proved that this was true, going from quartz
crystal microbalances with mass resolutions in the microgram range down to nanomechanical
devices, now with mass resolution in the octogram range, 10 to the minus 24 grams.
And in fact, our work with top-down nanofabricated devices, you know, basically has sort of peaked
back in the mid-2000s with zeptogram or slightly sub-zeptogram nanomechanical mass sensitivity,
and work that has been done in several places, but the first being in Andreas Bakthold's
group in Spain, showed that with a single low carbon nanotube, it could get down to
this 10 to the minus 24 grams.
Now this is, you know, sort of represents the truth in what we've been saying, that
you can get these 12 orders of magnitude improvement, and it looks like it can get down to single
AMU or single octogram mass resolution, but the next step is to apply this to some useful
technique.
So we know, and I'll come back to this again at the end, we know that macromolecules that
we want to measure are many times larger than the width of these devices, and so these are
not useful devices.
They're useful devices if you want to look at atomic species, you know, like low mass
atomic species, but if you want to do biological mass spectrometry, we need to stick with basically
CMOS compatible technologies, doing top down, and using widths that are roughly no smaller
than the size of the molecules themselves.
I'll come back to that.
So how do we turn, you know, this sort of nanophysics demonstration of, you know, hugely
sensitive mass sensing into mass spectrometry?
Well, just let me remind you what mass spectrometry is.
Mass spectrometry, we start with biological or chemical analytes in a liquid phase, and
we spray them into a high vacuum situation in such a way that ultimately the species
that are, the ionic species that are dissolved in solution, ultimately lose their solvent
phase, their aqueous phase, and by the time they get to the mass analyzer, are denuded
of all the solvent or aqueous phase, and they're essentially bare macromolecular ions.
These ions can then be manipulated in magnetic fields or electric fields, okay?
So we ionize the species that are in solution, we denude them of the solvent, and then we
can basically transport them into some sort of either electric or magnetic or combined
mass analyzer that separates them out into packets that are delayed in time that have
different mass to charge ratio, and these different packets of ions that have different
mass to charge ratio then are delivered at different times and we detect them as blips
in charge deposition onto a charge detector, okay?
And through that, and through something like 50 years of, excuse me, difficult, hard-won
technology to convert mass to charge ratio into biologically useful information, people
have developed this tool which is now the mainstay for proteomics, which is Proteus,
which you know are the machines, life's machines, basically.
So what's the difference between this previous paradigm and now mechanical mass spectrometry?
Well, then the mechanical mass spectrometry is actually quite a bit simpler conceptually.
We start with, again, preparing our species in the aqueous phase, injecting them into
vacuum.
Now, in this case, we don't actually need to have ionic species because we're not going
to detect charge.
We're going to detect the inertial mass of the species.
We're going to spray these onto a pixel array where each pixel is an individual mass detector.
It's going to register clicks with the incoming atoms, and then we have some sort of highly
multiplexed electronic readout.
Now, where is the mass analyzer?
Remember there was a mass analyzer that separated the spray of molecules into different mass
to charge ratios?
Well, we don't need that because our mass detector is the mass analyzer.
We can read out the frequency shift is directly proportional to the mass of the accreted species.
So NEMS acts as both the mass analyzer and mass detector.
Advantages are, as opposed to conventional mass spectrometry needing millions of molecules,
we can do single molecule detection.
Single Dalton resolution has been shown for atomic species, and we believe it's still
possible for molecular species.
We can measure extremely large masses.
This is key if you want to look at protein complexes, for example, big assembly of machines,
for example, ion channels and things like that.
I mean, there's many, many examples, ribosomes, for example, others.
These are, in some sense, on the scale of the forces that are involved in ionization
for conventional mass spectrometry.
Ionization and going into the fluid phase.
Typically, large species are weakly bound on sort of a Kaloubic scale and basically
break up.
And so it's difficult to do macromolecular complex mass spectrometry with the existing
paradigms of mass spectrometry.
Whereas we can do this with neutral species, and we can look at extremely large species
without loss of resolution.
In fact, we have more and more resolution as we go in mass.
We can make many measurements per pixel so that we can, if we actually put thousands
of mass spectrometers in a chip-based paradigm, we can actually get high throughput detection.
But of course, all of these things that take the initial demonstrations that we've done
with a single nanodevice and scaling it up to biological utility require all the tools
for semiconductor large-scale integration.
And to this end, over the past six years with collaborators in Grenoble at the equivalent
of the Department of Energy's Laboratory for Electronics and Information Technology
in Grenoble, also Minitech laboratory there, we've developed standard NEMS process.
We developed NEMS CMOS processes and 3D NEMS integration processes to allow this kind of
large-scale integration for these mass detectors to occur.
Now, a complication arises.
And we saw this basically in our first paper where we measured these single molecule jumps
for accretion is that, and this wasn't unexpected, that when we went to larger molecules,
it, this thing that I'm going to tell you about, became extremely manifest.
That is, even though we were sending in a single species, in this case beta amylase
of a few hundred kilodaltons, we got jumps of different height.
Now, the reason why we got jumps of different height is because there's a position dependence
of the mass sensitivity.
If you think about it, if you're operating this thing in fundamental mode, there's antinodes
here, sorry, there's nodes here and an antinode in the center.
So the maximum of the vibration is at the center.
And you would expect where the thing is vibrating the most, you actually have the biggest effect
on the inertial mass when you stick a little species on there.
So how do we get around that?
So how we get around that is, it turns out that each of the modes of the nanomechanical
device, you know, the first one being this, the second one being this, and so on, depend
differently on the position of the accreted species.
So in simplest form, we can actually, instead of just applying one control loop, we apply
two of these low noise control loops to simultaneously drive two modes, in this case the first and
the second mode, and read out, so here's the first and the second mode, and read out
the time-correlated jumps in the frequency that occur with the accretion of an individual
species onto these devices.
Notice that all the jumps occur in a time-correlated fashion.
They're happening to the same device, but the effect of two modes is different.
Okay, so now we have two measurements, and we have two equations for mass and position
unknowns, and we can basically take these two position-dependent frequency jump equations
and solve them to actually get both the position of the accretion and the mass of the species
in real time for each molecule that comes into the nanomechanical device.
So when I say we're making arrays of hundreds or thousands of mass pixels that are coupled
with CMOS electronics, each of these pixels has to be a multi-mode detection, right, because
we want to, when a molecule comes in, we want to capture it effectively.
There's a plume, right, that comes out from our spray.
We need to tile the area of that plume with good filling traction with lots of nanomechanical
devices.
Each one of them is a pixel that's read out independently with this multi-mode detection,
and that's what we've been building with our colleagues in France.
Now, what's interesting is that with this multi-mode detection, for part of his PhD
thesis, my former grad student, he's now a member of professional staff at Caltech,
basically understood that we could evaluate the analyte's uncertainty in mass and position
of adsorption with knowledge of basically the measurements of the noise sources in this
process.
So we can go from the experimental data, which is as a function of time, the time correlated
frequency jumps associated with these individual accretion events, to translating these things
into essentially, you know, here's the frequency jumps now, so the frequency jump, frequency
jump plane, and now it turns out that analytically we can calculate what the contours are of
constant mass and constant position, and then plot essentially out the probability
distribution function in the mass and position plane.
So each one of these is now like a little spectrum, if you will, both in position and
mass, you know, with uncertainty associated with the frequency uncertainty or the noise
uncertainty in our system.
Now, let me just take a, before I show you some recent data that we've gotten with this
system, it turns out that, let me tell you a little bit about where we've been going
with actually the front end of the system, trying to make it more and more amenable to
doing useful biological experiments.
The first thing that we used for injecting our protein species into these systems with
the nanomechanical device at the end was electrospray ionization, which is a standard process that's
used to get, in mass spectrometry, with a charged needle to essentially get protein
ions in solution to become protein, denuded protein ions in vacuum.
Okay, so usually there's, you inject in, you know, with an electrically charged needle
into the vacuum space, and then you go through differential stages of vacuum to progressively
evaporate away the solvent, and we transport this in sort of a coherent, you know, spatially
localized way by using radio frequency ion guides, hexapole ion guides, that ultimately
deliver the molecules down to the surface of our individual nanomechanical device, or
now into our nanomechanical arrays.
So the electrospray solution, droplets enter the capillary tube, okay, droplets in the
system here now undergo coulombic fission, ultimately only a single protein, single protein
ions remain, and then the proteins are brought down to where our nanomechanical device is.
In our first systems, you know, we were going from the era of making measurements at low
temperature, you know, in high magnetic fields, or mesoscopic physics, so, you know, I converted,
or I, you know, with my postdoc, converted one of our previous systems into our first
mass spec system, and our delivery distance was like a couple of meters, so you can imagine
our efficiency was really small.
These have now been converted to tabletop systems.
Now, what we've been going to lately is a variant of what's called MALDI, for those
of you who are familiar with mass spectrometry, which is matrix-assisted laser-induced, or
laser desorption ionization, laser desorption ionization.
We're going from there to sort of a new variant, which is called laser-induced acoustic
desorption, and the idea here is, on the front surface of a plate, you have the species,
or these can actually be individual cells that you're interested in analyzing, so either
macromolecular species, proteins, or individual cells on the front surface, and the back surface,
basically, you blast with a laser in a spatially accurate way.
This is a multilayer structure that's designed so that the optical radiation that hits on
the back is converted to a shock wave, an acoustic shock wave, and that propagates quickly
to the surface, and then blasts off the molecules, or the cells, into a little plume, and now
we put, at close proximity, our pixel array of mass sensors, okay?
And of course, we did this initially with a single mass sensor that's shown here.
So proteins are deposited on the substrate, and this is basically the same process that's
used for preparing MALDI plates today for conventional mass spectrometry.
The substrate's loaded into a UHV chamber, a pulse UV laser hits the substrate backside,
and then the shock waves happen.
Proteins dissolve, and then fizzies are brought to nearby coolings array, and we can detect
the mass, molecule by molecule.
So one of the first things we looked at are antibodies that occur in human blood serum.
These are called immunoglobulin M's, and immunoglobulin M's are sort of a Y-shaped
molecule, and important for the story that I'm going to tell you, and sort of leading
up to thinking about looking at protein complexes.
Um, these things in blood serum are most prevalent in a pentameric form.
They sort of associate into a biologically active pentameric form, but there are other
isoforms.
The pentamer is one isomer, one isoform, hexamer, tetamer, you know, a whole bunch of different
isoforms occur.
And in fact, these isoforms have different biological or immunological activity.
If you have zillions of molecules, you can run a gel electrophoresis process, and actually
see different bands in, under different conditions for these different isoforms.
Okay, again, this is having jillions of molecules.
But if you only have a few molecules, you can use Nernst mass spectrometry, and so this
shows 73 events for detection of this aqueous phase, this human serum IgM.
And the way this is built up is that here's the first event that happens.
Okay, the second event that happens is here, and we carry along for a ride in this representation
of the first event.
The third event happens here.
This is the first and the second event, and so on.
Right?
And so we can build up slowly, you know, molecule by molecule, the conventional spectrum that
you get for IgM in solution.
Okay, and you can see basically all these different prevalences based on the mass of
the species.
So that one megadalton is the tetameric form.
Now, usually all you get in mass spectrometry is the white trace, right?
But we get these individual units one by one, and so we know deterministically how this
white trace is actually composed.
Okay, again, usually you get the sum curve when you do mass spectrometry.
We get these individual units.
So this is in some sense a reversal of the traditional situation where you, you know,
in the traditional situation, you get the mass spectrum, you postulate what the individual
units look like, and then you use a multi-parameter fit to actually try to figure out what their
spectral weights are.
In this case, our measurement gives us directly the spectral weights.
We know for this particular experiment, we got precisely seven molecules of primer, you
know, 12 molecules of tetramer, 20 molecules of tantamer, and so on.
So this is sort of an exciting new development for actually looking at the prevalences in
quantitative way for these different species.
So where would we like to go next with this technology?
Well, remember I said that we're working with devices that are larger than nanotubes
that were forced, and that nanotubes give you single AMU resolution, which allows you
to look at protein fragments in the standard realm of biological mass spectrometry and
begin to figure out what's what.
Right now, with our mass resolution floor being something like 20 kilodaltons, our principal
area of application has been proteins, large proteins, protein complexes, viruses, and
as I'll tell you about some work with individual bacterial cells.
So what's interesting in this range that we can access is the first question.
How can we maximally, how can we maximize the efficiency and throughput?
So in this case, the question arises from the fact that all of the initial measurements
that we made were made with individual nanodevices, and if you think about the relative cross-sectional
area of the plumes that we inject, that we deliver, compared to the cross-sectional,
the captured cross-section of the individual nanodevices, it's infinitesimal compared to
the whole.
So our efficiency is really small.
It's like a part in 10 to the 9.
For every billion molecules we inject, we collect one.
I often get the question, how do you, how can you look at one by one, you know, how
fast do the molecules come?
Well, if you have a capture efficiency of 10 to the minus 9, the molecules come every
minute or something like that.
So, you know, why do we have traces with 73 molecules?
Well, that's about as long as we want to make the run to capture those 73 molecules.
Clearly, that's inefficient.
It's a demonstration.
We need to figure out how to get millions of molecules per second.
So what is the road map for actually doing that?
And we'll go into that as I'll show you.
And then finally, the question of, if we stick with devices that are large enough to be useful
biologically to capture wild molecules, not individual atomic species like nanotubes,
but things that have appreciable cross-sections, like maybe 50 nanometers or so, 100 nanometers
at the outside, can we push down the mass resolution to get a single AMU?
And back in this paper that we wrote in 2004, we convinced ourselves that that was possible.
Why haven't we achieved it?
So these are the three things that I'll talk to you about.
So targeted near-term applications first.
So I talked about macromolecular complexes.
And one of the things that we've been studying, and I won't go into the details of our results
so far, is things like ribosomes that are composed of many different protein subunits.
And it's interesting to actually look at, for example, samples that are taken from inside
the cell where one can look at the state of completely assembled ribosomes and actually
all sorts of other misformed composites and things.
So we can actually look at, on a single-cell level, what we begin to understand through
incompletely formed complexes.
What are some of the processes by which this macromolecular assembly is carried out?
Another example is that recently there have been advances in conventional mass spectrometry
where it's become possible to do fingerprint analysis on bacteria.
Not on individual bacteria, but on cultured bacteria where you have billions of bacteria
and you grind them into a homogenate.
And if they're all one and the same bacteria, you can get a mass spectrum fingerprint of
protein fragments.
And in this case, it's between 2 and 20 kilodaltons of protein fragments for the bacteria.
And it's been found that there are characteristic fingerprints, these mass spectra between 2
and 20 kilodaltons for bacteria, that are sufficient to identify bacterial species.
Now, this is interesting because bacterial identification is key for early-stage application
of antibiotics for these strongly resistant strains of bacteria that give people sepsis
in hospitals.
And it's a significant worry because if you apply all sorts of antibiotics before you
figure out what it is, and you've got to do it quickly before the person basically becomes
deceased, if you apply all sorts of antibiotics, basically you'll kill off all the other ones
except for the resistant one, and then that one will even more accelerate its deleterious
processes in the body.
So we really need to nail sepsis by having an early-stage diagnosis of what bacterium
it is, and then have a quick target for those specific bacteria.
Again, this is not something that we have today.
It's an extremely quick identification of bacteria.
So we'd like to take this idea of bacterial identification down to the single bacterium
scale.
We don't have to wait two days to do the culture because the patient might be dead,
but actually do it basically instantaneous in real time.
And so if we can translate our technique into a single bacterium model, that will actually
do the job.
I'll say a little bit more about that.
Another interesting thing in the vein of what I was talking about with ribosomes is actually
looking at assembly processes for viruses.
And just like chemical reactions, if one can actually look at all the variants of chemical
processes, things that actually don't go the right way, we can actually begin to understand
different processes that are taken in an assembly process.
And so we need to have a single-species resolution in order to do that.
How can we maximize efficiency and throughput?
How can we get from 10 to the minus 9 efficiency up to at least a few percent efficiency that
conventional mass spectrometry has, and perhaps even better?
Well, again, it's by tiling this area that we inject the plume with densely with NEMS
mass pixels.
And that's going to require large-scale integration of nanochemical devices.
So back in 2007 with our colleagues in Grenoble, we transferred our CMOS-compatible NEMS processes
in and into the foundry and worked out all the bugs in getting this to work on the 200-millimeter
8-inch wafer scale development of mechanical devices.
And then since then, obviously, we need per pixel a fair amount of electronics power.
We've co-integrated this with CMOS, and we're beginning to do 3D stacking runs to figure
out how to do this 3D integration.
And the reason for the 3D integration is perhaps you know how the Texas Instrument
Micromirror works for digital projectors.
It's basically a dense tiling of little torsional devices.
And these devices have densely tiled, so you have good optical coverage.
And below them is all the electronics that drives the individual pixels.
And so we're essentially attempting to do at the nanoscale the same sort of things with
these multimodal mass sensors.
So what about trying to get down from 20 kilodaltons and above, you know, this realm that we can
study protein complexes and large protein macromolecules down into the few-dalton range,
which is the province of looking at protein fragments and allowing us to do quantitative
mass spectrometry.
How can we actually get there?
And again, you know, the first thing to say is that we need devices with the size of macromolecules.
So we can't take a shortcut and use nanotubes to get there.
We need to stick with the CMOS compatible, again for the large-scale integration aspect,
but also because of the size needing to be comparable to the size of macromolecules.
So what are prospects for improving mass resolution if we restrict ourselves to the VLSI platform
and talk down NEMS devices that are 50 or 100 nanometers in size?
Well, the first thing is, how do we get our minimum resolution down from, you know, 10
or 20 kilodaltons, you know, to a few A of U?
I mentioned that the first paper that we wrote said that would be technically feasible, and
we've spent about 10 years exploring how close we can get to fundamental thermodynamic limits
with this mass resolution.
And the bottom line is that there right now is what we perceive to be an engineering brick
wall.
There are noise fluctuation processes that we don't completely understand that are probably
surface-mediated that we've been exploring, but this is complicated, difficult surface
science that needs to be applied to nanomechanical devices.
And so we're continuing to investigate that.
Now, in our defense, I'll say, if defense is needed, that even in large macro devices,
we don't really understand in great detail at the microscopic level what noise, what's
the origin of noise processes.
And so maybe it's, you know, a bit arrogant or brash to think that in a short period of
time we'll understand these for nanomechanical devices, but you can see the reason why this
is really imperative for us.
So we continue to push on that.
We believe that we have advances in hand to allow us to go down to, you know, something
like several hundred kilodaltons within the next couple of years.
Now, a second thing that actually goes into the mass sensitivity is basically how hard
you actually drive the devices.
Those of you that know about making oscillators, you know that you get the best phase noise
performance, the purest tones out of oscillators, if you drive them extremely hard, all the
way up into some sort of natural limit.
And it turns out that that natural limit of driving is typically the nonlinear regime.
Once you go beyond the non, once you go into nonlinearity, you start making harmonics and
the amplitude of the system depends on the frequency that you're at.
And so if you have frequency fluctuation, that's converted to amplitude fluctuations
and vice versa.
And so we've looked at this very carefully.
And instead of just assuming that this conventional paradigm is true, namely that if you want
to make good oscillators, which is what we're doing with these pixels, if you want to make
good oscillators, you keep them in the linear regime.
We've looked at that and we've found that in fact, you can go strongly into the nonlinear
regime where you have this amplitude and phase noise conversion.
But we can find special, this is a paper that just came out in physical letters.
We can find with these nanomechanical oscillators that are based on nonlinear resonators, we
can find special places in the operating curve where in fact, this amplitude, the frequency
noise conversion is suppressed.
So this is another way that we're exploring.
And this in part will help us to actually push down the ultimate line submission.
So let me just conclude with a couple of slides.
And I think there will be enough time for some questions if people have them.
Part of our roadmap right now is trying to push this and see if we can do this single
bacterium identification for mass spectrometry.
But what we'd really love to do is using microfluidic systems that are well developed
and you know, many great examples from laboratories here of single cell handling and microfluidics,
to be able to do mass spectrometry on a cell by cell level to understand, to take snapshots
instantaneously of what all the machines in the cell actually do.
And just to give you some sort of numbers for this, with 10,000 analyzers in parallel,
which is, you know, a medium scale integration feat that we think we should be able to do.
With a protein flux that we can handle now of about per pixel of about 1,000 molecules
per second, we should be able in less than two minutes to measure essentially all the
proteins, you know, the billion proteins in a cell, and then go through many, many cells
with, let's say, you know, arrays of these, arrays of mass spectrometers that are individual
boxes made into some sort of a screen laboratory.
I was going to say something about imaging mass spectrometry, but I think I'll skip
that in the interest of allowing people to ask questions.
So we've been pushing recently towards, you know, after a bunch of years of just sort
of doing the physics of mass sensing and understanding noise in that mechanical system.
Since about 2006, we've been really pushing to try to develop systems using these for
doing biologically interesting mass spectrometry.
And it's going to take us a few more years to get where we want to go.
But I think it's an interesting and worthy enterprise for actually taking these nanodevices
and turning them into nanosystems.
And here's all the great people that I've been privileged to work with on this project
and some of the funding that's actually supported this.
Thank you for inviting me to speak today.
As before, we're going to take a few questions.
But first, I'd like to present Michael with this plaque and our appreciation for coming
out here to visit us and present this keynote address.
Thanks very much.
Thank you very much.
We have time for some questions.
You mentioned about the use of nonlinearity.
Have you tried parametric configurations?
So we have.
Parametric...
So we've done an extensive amount of work with parametric mechanical systems.
And parametric processes don't improve the fundamental signal-to-noise ratio that you
have.
In going from the mechanical domain to the electrical domain, they can basically boost
the displacement amplitude of the system and allow you...
So it doesn't improve the signal-to-noise at the front end, but if your signal-to-noise
in large part is limited by displacement transducer noise, it can give you big displacement
to overcome that noise.
So it turns out in these systems, our displacement transducer noise is not the principal thing
in this.
It's basically this anomalous phase noise that I was alluding to in the mechanical domain.
And for that, we need to get into the mechanical...
We need to do all of our engineering in the mechanical domain to really understand what
the origin in the mechanical domain is of these processes.
And we can talk more about parametric stuff, but with the specific context of this discussion
of parametric spectrometry, parametric won't really help us because we've already done
a good job of engineering the noise in our transducers.
So moving toward your single-cell models, how do you do your sample processing in terms
of salts and fuel salts?
So this is work that's in progress.
And we haven't worked out all the details about what, in terms of the salts, are important
in actually getting these mass spectra.
But all the sample preparation that we've been using...
So we have partners from DNRU, which is a big pharmaceutical company, sort of in parallel
with Davidson in the U.S., that does these biological fingerprinting using mass spectrometry.
And we're using the same protocols that they have.
That's not really an answer that I can tell you about.
What's your...
Being intentionally basic.
Full disclosure.
What's your experience with functionalizing these?
And do you think that could improve efficiency?
So remember that we're doing everything in the vacuum phase.
And molecular recognition in the fluid phase is based on secondary, tertiary, paternary
structural proteins.
And as soon as you absorb them onto the surface of the device, if it's in fluid, you can preserve
much of that three-dimensional structure and preserve the molecular recognition walk-in
key process.
But in vacuum, the proteins denature and you lose all that.
So you can't make these biologically specific because we're working in the vacuum phase.
That's why we rely on either microfluidic preprocessing.
You can use immunospecific microfluidic preprocessing upstream from the detection in the same way
that you do cell B and other things in mass spectrometry.
Or at the downstream end of that, at the ends themselves, we look at exactly the half the
inertial mass of the species as our discriminator.
So I know that your group has also done nanomechanical resonators, but nanofluidic channels built in.
No, we haven't.
Oh, you haven't?
No, we've been collaborating with Scott Manalis, but we haven't ourselves done that.
Right.
I saw a paper and a letter.
Yeah.
So is there any, what are the limitations of that technology?
Yeah.
But ultimately.
So you could do what was previously described about using immunospecific detection in small
channels for fluidic detection with high quality factors.
So for those who are not fully apprised of what's being alluded to here, when you take,
the reason why we get these 12 orders of magnitude improvements is that we stay in vacuum.
We preserve high Q for the devices.
And the question is, if we take these into liquid phase where we can do some of this
immunospecific molecular recognition for specific species directly in the liquid phase, how
well will they work?
But we still get this incredible mass sensitivity.
Well, what happens is that they get fluid gaps.
The quality factor goes down.
You get four or five orders of magnitude decrease in the mass sensitivity.
So our colleague, Scott Manalis and his group at MIT said, well, let's not put the devices
in solution.
Let's put the solution in the devices and keep the devices in vacuum with the fluid
channels inside.
And then you can do all the fluidic stuff and do these little micro and half channels.
So what we've done, our contribution, is actually looking at how far you can scale these fluid
embedded nano channels, their sensitivity for immunospecific detection.
And the big problem comes, so they end up doing pretty well, comparable to other sensors.
But because of a whole bunch of technical difficulties, you never really get the full
promise of what we can do in vacuum.
And one of the big limitations turns out to be that the throughput of these little nano
channels means that unless you make arrays of thousands of these individual nano channel
resonators, you don't get sufficient throughput to look at biologically useful quantities
of fluid to do analyses.
When you're talking about, for example, looking at HIV detection and trying to do it without
PCR, for example, you're up against a really stringent criterion of only looking for a
couple of molecules per milliliter.
And then you think about nonspecific binding and how much fluid you have to push through
in a milliliter.
But it becomes prohibitive to actually, even though it's in principle feasible, practically
there's a whole bunch of reasons why we can't do it.
We should probably move on, but we won't let him leave town until you have a chance
to talk to him at lunchtime.
Let's thank our speaker again.
Thank you.

