English Research Yasuyuki OZEKI's personal web

(1)

High-sensitivity

stimulated Raman scattering

(SRS) microscopy

1Graduate School of Engineering, Osaka University

2 _JST-PRESTO

Yasuyuki OZEKI^1,2 and Kazuyoshi ITOH¹

Outline

• Stimulated Raman scattering microscopy

• Comparison with previous Raman

microscopy

• Theoretical sensitivity limit

• Sensitivity improvement by subharmonic

synchronization technique

Thank you for introduction. Hi, I’m Yasuyuki OZEKI from Osaka University, Japan. Today, I’d like to talk about the stimulated Raman scattering, or SRS microscopy, which is an emerging alternative to CARS microscopy.

I’d like to introduce the basic ideas of SRS microscopy and present our recent system with high sensitivity.

Here is the outline of my talk. First, I’ll introduce the SRS microscopy and compare it with previous Raman microscopy to highlight the feature of SRS microscopy. Then I’ll discuss the theoretical sensitivity limit, referring the principle of SRS. Finally, to achieve the sensitivity limit, I’ll demonstrate our high-sensitivity system.

(2)

Outline

• Stimulated Raman scattering microscopy

• Comparison with previous Raman

microscopy

• Theoretical sensitivity limit

• Sensitivity improvement by subharmonic

synchronization technique

Stimulated Raman scattering

Wavelength conversion [1], laser oscillation [2], laser spectroscopy [3-5], optical amplification [6], wavelength tunable pulse source [7],

optical microscopy [8-12]

[1] Eckhardt et al., PRL 9, 455 (1962) [2] Takuma and Jennings, APL 4, 185 (1964) [3] Jones and Stoicheff, PRL 13, 657 (1964) [4] Owyong and Jones, OL 1, 152 (1977) [5] Levine et al., JQE 15, 1418 (1979) [6] Namiki and Emori, JSTQE 7, 3 (2001)

[7] Nishizawa and Goto, PTL 11, 325 (1999) [8] Nandakumar et al., FRISNO-8, Mo-B (2005) [9] Ploetz et al., Appl. Phys. B 87, 389 (2007) [10] Freudiger et al., Science 322, 1857 (2008) [11] Ozeki et al., Opt. Express 17, 3651 (2009) [12] Nandakumar et al., New J. Phys. 11, 033026 (2009)

ω_R Resonance: ωR = ω1 − ω2

Molecular vibration

ω2 ω1

E(ω)

ω Increase Output field

ω2 ω1

E(ω)

ω Input field

ω1

ω2

Applications

Decrease

So, what is SRS microscopy?

Stimulated Raman scattering is a well-known nonlinear light-matter interaction between two-color light and molecular vibration. When the optical frequency difference matches the molecular vibrational frequency, the photon energy is transferred from the high-frequency side to the low-frequency side. This is SRS. SRS has a variety of applications such as wavelength conversion, laser oscillation, laser spectroscopy, optical ampliﬁcation in optical ﬁber communication systems, and wavelength tunable pulse source. Recently, SRS has been applied to optical microscopy.

(3)

SRS microscopy

OB OB

Sample ω₁

IM ω₂

Lock-in

@ f_m t

ΔI_SRS t

f_m _t

[1] Nandakumar et al., FRISNO-8, Mo-B (2005) (Stuttgart Univ., Dr. Volkmer’s group) [2] Freudiger et al., Science 322, 1857 (2008) (Harvard Univ., Prof. Sunny Xie’s group) [3] Ozeki et al., Opt. Express 17, 3651 (2009) (Osaka Univ., Prof. Itoh’s group) [4] Nandakumar et al., New J. Phys. 11, 033026 (2009)

•

Label-free imaging

•

^3D

^•

High sensitivity

•

High contrast

Nature, News Feature, vol. 459, p. 636, June 2009

That is SRS microscopy. What you see here is the basic conﬁguration of SRS microscopy. We use two-color laser pulses, one of which is intensity-modulated in time. Then the pulses are combined and focused on a sample. Through the SRS process, the intensity modulation is transferred to the other pulse. The transferred modulation is detected by lock-in detection technique, so that we can detect the amount of a speciﬁc molecular vibration. Images are obtained by scanning the sample position or the focus position.

The features of SRS microscopy include label-free imaging, three-dimensional resolution, high sensitivity and high contrast.

The SRS microscopy was ﬁrst suggested by Dr. Volkmer’s group in Stuttgart University, and recently biological imaging was demonstrated independently in Harvard University, Osaka University, and Stuttgart University.

Last year, SRS microscopy was introduced in the Nature journal although almost all the article describes the achievements by Harvard group. These are cellurose and lignin in plant cells, water, protein and oil in a soya drink, and lipids in brain

tumours.

(4)

SRS image of HeLa cell

Cultured HeLa cell, Raman shift: 2850 cm^-1 (CH2 stretch), 40 x 40 x 10 µm² Optical power: 1.9 mW & 1.9 mW, pixel dwell time: 0.5 ms

3D SRS image of HeLa cell

Cultured HeLa cell, Raman shift: 2850 cm^-1 (CH2 stretch), 40 x 40 x 10 µm² Optical power: 1.9 mW & 1.9 mW, pixel dwell time: 0.5 ms

We are trying to observe live cells as well. This is a cultured HeLa cell. The Raman shift was set to 2850 cm-1, CH2 stretching mode, to visualize the distribution of lipids.

SRS microscopy has three dimensional resolution because the SRS process occurs only at the focus. Therefore, we can observe the sample in 3D from various directions.

(5)

3D SRS image of plant cell

Cultured tobacco BY-2 cell, Raman shift: 2850 cm^-1 (CH2 stretch), 60 x 60 x 40 µm² Optical power: 2.0 mW & 1.3 mW, pixel dwell time: 0.3 ms

Outline

• Stimulated Raman scattering microscopy

• Lock-in of SRS --> Label-free imaging

• Comparison with previous Raman

microscopy

• Theoretical sensitivity limit

• Sensitivity improvement by subharmonic

synchronization technique

This is a plant cell, showing the three-dimensional distributions of various cell components such as cell wall and nucleus as well as several droplets, which are likely assigned to mitochondria.

In this way, SRS microscopy exploits the lock-in detection of SRS for label-free imaging.

(6)

Outline

• Stimulated Raman scattering microscopy

• Comparison with previous Raman

microscopy

• Theoretical sensitivity limit

• Sensitivity improvement by subharmonic

synchronization technique

Raman microscopy

Spectral analysis Raman scattering

Raman spectrum

‘molecular fingerprint’

G. J. Thomas, Jr., Ann. Rev. Biophys. Biomol. Struct. 28, 1 (1999)

DNA Protein

Lipid

Raman scattering microscopy for ‘label-free imaging’

Hamada et al., J. Biomed. Opt. 13, 044027 (2008)

Issues weak signal slow acquisition Molecular

vibration @ ωR

ω

(ω − ωR)

So, how can we compare it with previous Raman microscopy?

As all of you may know, Raman microscopy has been attracting considerable attention. Raman scattering is the inelastic scattering of photon by molecular vibration. The scattered light is spectrally analyzed to get Raman spectrum, which has rich spectroscopic information of the sample. Therefore, Raman scattering microscopy allows label-free imaging. This is a Raman image of a HeLa cell reported by another group in Osaka University. They used a confocal Raman microscope with line scanning conﬁguration. They succeeded in visualizing cytochrome c, protein, and lipid without labeling.

An important issue of Raman microscopy is that the signal is weak, requiring long acquisition time.

(7)

Issue

Nonresonant background

CARS microscopy

coherent anti-Stokes Raman scattering

Resonance: ω₁ − ω₂ = ω_R Optical

pulses

ω₁, ω₂ 2ω₁− ω₂

M. D. Duncan et al., Opt. Lett. 7 (1982) 350. A. Zumbusch et al., Phys. Rev. Lett. 82 (1999) 4142. M. Hashimoto et al., Opt. Lett. 25 (2000) 1768.

Features Strong signal

Fast acquisition (up to video rate) CARS signal

NIH3T3 cell, 1579 cm^-1

Volkmer, J. Phys. D, 38 (2005) R59

Four-wave mixing ω1

ω2

(2ω1 − ω2)

Low contrast Spectral distortion

Nonresonant background

Intensity: I_CARS ∝ |χ⁽³⁾|²

Origin: Re χ

⁽³⁾

(χ⁽³⁾: 3rd order nonlinear susceptibility)

Electric field_{: E}_CARS ∝ χ⁽³⁾

ω₁− ω₂ ω_R

χe⁽³⁾

Re χR⁽³⁾

Im χR⁽³⁾

χ⁽³⁾

R. W. Boyd, Nonlinear Optics.

|χ⁽³⁾|²

（electronic nonlinearity_） Spontaneous Raman: I_Raman ∝ Im χ⁽³⁾

Susceptible to Re & Im

The sensitivity issue was effectively solved by introducing CARS microscopy, as all of you certainly know. In CARS microscopy, we launch two-color laser pulses, which strongly drive molecular vibration, resulting in strong four-wave mixing signal, which we call CARS signal. The CARS signal is so strong that fast acquisition up to the video rate is possible. An important issue of CARS is nonresonant background, which reduces the image contrast and distort the vibrational spectrum.

Here I’d like to explain the nonresonant background. Mathematically, the

nonresonant background originates from the real part of χ⁽³⁾, which is 3rd order nonlinear susceptibility. Since the electric ﬁeld and the intensity of CARS signal are proportional to χ⁽³⁾ and its squared modulus, respectively, CARS is susceptible to both real and imaginary components. What you see here is χ⁽³⁾ as a function of difference frequency of two-color light. An important thing is that the imaginary part exists only around the vibrational resonance, whereas the real part exists in all the frequency, due to the electronic nonlinearity. As a result, the square modulus has an offset with asymmetric line proﬁle. This is the origin of nonresonant background. On the other hand, since the spontaneous Raman scattering signal is proportional to the imaginary part of χ⁽³⁾, it is immune to nonresonant background.

(8)

Multiplex CARS + spectral analysis

Suppression of

nonresonant signal

λ/2 Pol.

ωp

ω_s

J. Cheng et al., OL 26 (2001) 1341.

F. Ganikhanov et al., OL 31 (2006) 1872. J. P. Ogilvie et al., OL 31 (2001) 480. E. O. Potma et al., OL 31 (2006) 241. A. Volkmer et al., OL 31 (2001) 480.

ωp, ωs APD

ω_p1 _ω

p2 Optical SW

ω_s

ωp, ωs

H. Kano et al., OE 13 (2005) 1322. Spectro-

meter ωp, ωs

APD PCF

PMT

FM-CARS

PMT

Epi-CARS Polarization CARS

FT-CARS ωp, ωs

Interferometric CARS

Development of CARS microscopy ~ Mitigation of nonresonant signal

...leading to complexity, signal loss, slow speed...

...and more

Stimulated Raman scattering

Detection of tiny intensity change (~10^-4) Deteriorated by laser noise

Issue of SRS: low sensitivity??

CARS

2 µm

SRS

2 µm

Water

4 µm Polystyrene bead ω_R

Resonance: ωR = ω1 − ω2

Molecular vibration

ω2 ω1

E(ω)

ω2 ω1

E(ω)

ω Input field

ω1

ω2

ω2 Decrease

Reflects only Im χ⁽³⁾ Immune to Re χ⁽³⁾

Background free High contrast

In CARS microscopy, it is crucial to suppress the nonresonant background. So far, various techniques have been proposed to suppress the nonresonant background. We may say that the development of CARS microscopy was almost the mitigation of nonresonant background. However, these techniques suffered from increased complexity, signal loss, and slow acquisition.

Stimulated Raman scattering is also the third-order nonlinear-optical effect. SRS reﬂects only imaginary χ⁽³⁾ and is immune to real χ⁽³⁾. Therefore, you can easily imagine that SRS provides background free images with high contrast. What you see here are SRS and CARS images of a polystyrene bead in water. SRS image has no background from water, leading to high contrast.

An important issue of SRS is its sensitivity. Since the detection of SRS signal requires the detection of tiny intensity change, SRS signal may be deteriorated by laser noise.

(9)

Comparison of Raman microscopy

ωS

ωCARS

Technique

Spontaneous Raman

CARS

SRS

Sensitivity Nonresonant background

Low

High

No

Yes

ωL ω

ω2

ω

ω2

ω1 ω ω1

High?? No

Outline

• Stimulated Raman scattering microscopy

• Comparison with previous Raman

microscopy

• No background, high sensitivity?

• Theoretical sensitivity limit

• Sensitivity improvement by subharmonic

synchronization technique

To summarize the features of Raman microscopy techniques, SRS has no nonresonant background, and this allows us to obtain high contrast images. Therefore, the key is whether it is possible to obtain high sensitivity with SRS microscopy.

(10)

Outline

• Stimulated Raman scattering microscopy

• Comparison with previous Raman

microscopy

• No background, high sensitivity?

• Theoretical sensitivity limit

• Sensitivity improvement by subharmonic

synchronization technique

Sensitivity limit of SRS

is similar to CARS

Y. Ozeki et al., Opt. Express 17, 3651 (2009).

SRS microscopy can be as sensitive as CARS

at the theoretical limit

To clarify this point, we considered the theoretical sensitivity limit of SRS microscopy.

Actually, the sensitivity limit of SRS is similar to CARS. This point is expressed by this equation. Therefore, SRS microscopy can be as sensitive as CARS provided that we could achieve the theoretical sensitivity limit.

(11)

Mechanisms of SRS & CARS

Classical picture

Input

ω_R

ω₂ ω₁ I(ω)

ω

Molecular vibration

Quantum-mechanical picture

ω₁ ^ω² ^ω¹ ^CARS _ω₁ ^ω² ^ω² SRS

Sensitivity comparison is easier

ω2

ω1

Output

ω2 ω1

I(ω)

ω SRS SRS

CARS

Principle of SRS

Force on molecules Refractive index modulation

ωR Resonance: ωR = ω1 − ω2

Molecular vibration

ω2 ω1

E(ω)

ω2 ω1

E(ω)

ω Input field

ω1

ω2

ω2 Decrease

In order to understand the sensitivity issue, we have to consider the mechanisms of SRS and CARS. They are often described in the quantum-mechanical picture. Instead, we introduce the classical picture, with which sensitivity comparison is easier.

In the classical picture, SRS is viewed as the excitation of molecular vibration, which results in the refractive index modulation. So, let’s look into the details.

(12)

ω_R Resonance: ω_R= ω1 − ω2

Molecular vibration

ω2 ω1

E(ω)

ω2 ω1

E(ω)

ω Input field

ω1

ω2

ω2 Decrease

Principle of SRS

t Intensity

Beat frequency

ω₁ – ω₂ Phase modulation

Refractive index modulation depending on q

(Doppler shift) Internuclear distance q

q ∝ Re χ⁽³⁾ cos(ω₁ – ω₂)t t

In-phase vibration

Lagged vibration due to resonance + Im χ⁽³⁾ sin(ω₁ – ω₂)t

Principle of SRS (spectral domain)

Im ω₂ ω₁ Re

ω Input field

ω Im

Re

Output field Phase modulation

spectrum

ω ω₁− ω₂

0

∝ Im χ⁽³⁾

∝ Re χ⁽³⁾

Im Re

iϕ(t) ∝ iq ∝ iRe χ⁽³⁾ cos(ω₁ – ω₂)t + iIm χ⁽³⁾ sin(ω₁ – ω₂)t

∝ (iRe χ⁽³⁾ − Im χ⁽³⁾) exp[–i(ω₁ – ω₂)t] (positive frequency) + (iRe χ⁽³⁾ + Im χ⁽³⁾) exp[i(ω₁ – ω₂)t] (negative frequency)

Fourier transform of expiϕ(t) ~ 1 + iϕ(t) (ϕ(t): optical phase)

The two-color pulses form an intensity beat at the difference frequency, which resonantly pushes molecules. As a result, the molecules start vibration at the same frequency. Therefore, internuclear distance q can be described in this form, which consists of an in-phase vibration and a vibration with a phase lag due to vibrational resonance. They are proportional to the real part and imaginary part of the

nonlinear susceptibility. This vibration leads to refractive index modulation, which leads to the optical phase modulation or Doppler shift. In this way, real χ⁽³⁾ and imaginary χ⁽³⁾ have different modulation phase of optical phase modulation. Furthermore, electronic nonlinearity causes intensity-dependent refractive index through optical Kerr effect and therefore contributes to Re χ⁽³⁾.

In the spectral domain, optical phase modulation can be described as the convolution of input ﬁeld and phase modulation spectrum, which is the Fourier transform of phase modulation in time. Since Real and Imaginary χ⁽³⁾ have different modulation phase of optical phase modulation, they have different phase

modulation spectrum. Importantly, the real part of the phase modulation sidebands are proportional to imaginary χ⁽³⁾, and their sign is negative and positive for positive and negative frequencies.

(13)

Phase modulation spectrum

Principle of SRS (spectral domain)

SRS: Interference between

the input field and PM sideband ^Homodyne

Photon counting Shot-noise limited sensitivities

of SRS and CARS are similar!

Im ω2 ω1

Re

ω _ω

2ω₂− ω₁ 2ω₁− ω₂ ω₂ω₁

SRS

Im CARS Re

射出光電場

SRS

ω ω₁− ω₂

0

∝ Im χ⁽³⁾

∝ Re χ⁽³⁾

Im Re

Input field Output field

CARS: PM sideband

Light applies force on

molecules

q

Polarizability

(microscopic dielectric constant)

Energy of polarization

Force F

W ₌¹ 2^{α E}

2 _{F =}^dW

dq ⁼ 1 2

dα dq ^E

2

Force on q

Similarity Electric

field E

C = C(x)

E-field E = V/d

Energy of

polarization ^{Force on x}

α = α(q)

Capacitance

F₌^dW dx ⁼

1 2

dC dx^V W ₌¹ 2

2^CV

2 ^χx

χx + L εr

x F

(εr = 1 + χ)

Through the convolution, the input ﬁelds interfere with the modulation sideband, changing their amplitude according to the imaginary part of χ⁽³⁾. This is SRS. Therefore, SRS can be viewed as the homodyne detection of a sideband from the imaginary part of χ⁽³⁾ by the excitation light. On the other hand, this sideband

corresponds to CARS, which is detected through a photon counting manner. So the origins of SRS and CARS are the sideband of the phase modulation spectrum. Here, considering that the shot-noise limited sensitivities of homodyne and photon counting are similar, we can conclude that SRS can be as sensitive as CARS.

Here, let me reinforce the foregoing explanations. How does light apply a force on molecules? When the polarizability is a function of the intermolecular distance, a force appears on q so that the polarization can increase its energy.

This situation is similar to a parallel-plate capacitor with a dielectric slab partially inserted. Since the capacitance is a function of the position x, a force appear so that the energy of the polarization increases. In this way, light can apply a force on molecules.

(14)

Phase shift at

resonance

Low-frequency

Resonant frequency

t

F F

F F F

t

t ForceF, Position x

Force F, Position x Power P = F(dx/dt)

Power P = F(dx/dt) Ave. = 0

Ave. > 0 In-phase

90 deg. phase lag

Impulse response & freq. response

of a damped oscillator

t h(t) = exp(−Γt)sin ω0t

t sin ω0t = i(exp−iω0t − exp iω0t)/2

t exp(−Γt)

ω Re Im

0 ω0

−ω0

ω Re Im

0

Phase lag due to resonance

Time domain Frequency domain

ω Re

Im

ω0

0

I told you that the amplitude and phase are dependent on the frequency of the force. This point can be understood by considering a pendulum or a swing. When you push slowly a pendulum, it moves according to the force. Thus the force and the position are in phase, and the average work is zero.

On the other hand, when you push the pendulum at the resonant frequency, the pendulum passes the origin when the force is maximum. Therefore, the position has a phase lag of 90 deg. with respect to the force, and the average power is positive.

More rigorously, we consider a pendulum as a linear, time stationary system, whose dynamics can be completely described by an impulse response. When an

impulsive force is applied to a pendulum, it gets a velocity, and then oscillates with damping. Therefore the impulse response is the product of exponential decay and sine curve. Its Fourier transform is the frequency response. Since a damped oscillation is the product of these functions, its Fourier transform is the convolution of the Fourier transform of each components. The Fourier transform of exponential decay is a Lorentz function, and that of sine curve has delta functions with

imaginary amplitudes at the oscillation frequency. In this way, Lorentz function is convoluted to the imaginary part, resulting in this frequency response. You can see that the frequency response is imaginary at the resonance frequency, which

indicates the phase lag due to resonance.

(15)

Shot noise basics

Photon picture: # of photon obeys Poissonian statistics

Prob. density

# of photon Ave.: n₀

Std. Dev.: n₀^1/2

Shot noise

Electric field picture: Vacuum fluctuation gives uncertainty of light energy

Im E

Re E Vacuum fluctuation

ΔE E _ΔE_I ΔEQ

(1) (2)

(2) → (1) _{Shot noise}

Photon

counting

Detector

E(t) _{i(t) ∝ n}

∝|E(t)|²

E(t) i(t) ∝

|E(t) + E_LO(t)|² E_LO(t)

Im

Re

Im

Re Vacuum

fluctuation

E_LO E Energy

hν(n + 1/2)

vs.

Shot-noise limited SNR: (signal field)/(vacuum field)

Homodyne

(CARS) ^(SRS)

I also introduce the basics of shot noise. The shot noise originates from the Poissonian statistics of the number of photon. Alternatively, we can consider that shot noise is caused by vacuum ﬂuctuation, which gives uncertainty of light energy. The both pictures give the same amount of shot noise.

The latter picture allows us to compare the photon counting with homodyne. In photon counting, we detect photon energy. In homodyne, we detect optical

interference. In both cases, the shot noise originates from the vacuum fluctuation. Therefore, shot noise limited SNR is related to the ratio of signal field and vacuum filed. This is why CARS and SRS have similar shot noise limited sensitivity.

(16)

R. W. Boyd, Nonlinear Optics.

• # of photon emitted at ω2

P = Dm₁(m₂ + 1)f_repτ.

Stimulated Raman Spontaneous Raman

ω₁

Vacuum field at ω₂ Spontaneous emission at ω₂ ω₁ ^ω²

Stimulated emission at ω₂

D: Proportional constant, m1, m2: # of photon per pulse, frep: rep rate, τ: averaging time

• SNRSRS^{= (P}st^{/ (m}2^frep^τ)^1/2⁾²^{, SNR}Raman^{= (P}sp^/Psp^1/2⁾²^{= P}sp

SNRSRS / SNRRaman = (SNRSRSm2/frepτ)^1/2 ~ 1000

SNRSRS: ~ 25 dB, frep: 76 MHz, 1 mW, τ: 1 ms

• Sensitivity merit of SRS over spontaneous Raman:

Y. Ozeki and K Itoh, Laser Phys. 20, 1114 (2010)

SRS ^vs. Spontaneous

Raman

Outline

• Stimulated Raman scattering microscopy

• Comparison with previous Raman

microscopy

• No background, high sensitivity

• Theoretical sensitivity limit

• Similar to CARS

• x1000 spontaneous Raman

• Sensitivity improvement by subharmonic

synchronization technique

Additionally, quantum-mechanical picture is useful for comparing the sensitivity of SRS with that of spontaneous Raman scattering. In the quantum mechanical picture, stimulated Raman is viewed as the stimulated emission, and spontaneous Raman scattering is viewed as the spontaneous emission caused by the vacuum ﬂuctuation. Therefore, the number of photon emitted is denoted by this equation. Then we can formulate the SNRs of stimulated Raman and spontaneous Raman as these equations. From these, we can denote the SNR merit of SRS over

spontaneous Raman is described by this equation, which depends on the SNR of SRS, repetition rate, optical power and averaging time. In the current experimental conditions, the SNR merit is as much as 1000.

In this way, the theoretical sensitivity limit of SRS is similar to CARS and 3 orders of magnitude higher than spontaneous Raman.

(17)

Outline

• Stimulated Raman scattering microscopy

• Comparison with previous Raman

microscopy

• Theoretical sensitivity limit

• Similar to CARS

• x1000 spontaneous Raman

• Sensitivity improvement by subharmonic

synchronization technique

Key for improving sensitivity:

high-frequency lock-in

10 MHz [2]

Popt: 4.5 mW

Tdwell: 100 ms ^PTdwell^opt^{: 1 mW}: 3 ms

100 kHz

Popt: 5 mW

Tdwell: 50 ms ^PT^optdwell^{: 0.6 mW}: 2 ms Popt: 10 mW

Tdwell: 200 ms

2 MHz [1]

Polystyrene beadsPlant cells

N.A.

[1] F. Dake et al., OPJ2008 (2008). [2] Y. Ozeki et al., Opt. Express 17, 3651 (2009).

So, how to achieve the sensitivity limit?

The key for improving the sensitivity is high-frequency lock-in detection. As I mentioned before, SRS signal is detected by the lock-in technique. What you see here is SRS images of polystyrene beads and plant cells, obtained with various lock-in frequencies 100 kHz, 2 MHz, and 10 MHz. You can see that the images become clearer as the lock-in frequency increases.

(18)

Photo-current spectrum

log freq. PD signal [log]

Shot noise

1/f laser intensity

noise

SRS signal SNR

f_m

Photo-current spectrum

SNR

f_m Shot

noise 1/f laser intensity

noise

Such a dependence on lock-in frequency can be understood by considering the frequency spectrum of the photodiode signal. The photodiode signal has a large 1/f noise due to the intensity noise of the laser and white shot noise. SRS signal

appears at the lock-in frequency, and this part corresponds to the SNR. Thus by increasing the lock-in frequency, we can reduce the 1/f noise and improve the SNR. If the sensitivity is still limited by the laser noise, it is worthwhile to further increase the lock-in frequency to push the sensitivity to the shot-noise limit.

(19)

Photo-current spectrum

Shot-noise limited

SNR

f_m Shot

noise 1/f laser intensity

noise

Toward higher lock-in

frequency

Previous configuration

No modulator & Maximum lock-in frequency Subharmonically synchronized pulses

BW limitation

However, in our previous conﬁguration, it was difﬁcult to further increase the lock-in frequency due to the bandwidth limitation of the modulator. To cope with this, we present a simple approach to increase the lock-in frequency. We use two-color laser pulses, one of which has a half of the repetition rate of the other. By doing this, SRS turns on and off alternatively. Therefore SRS is detected at the maximum lock-in frequency, allowing us to have the shot-noise limited sensitivity.

(20)

Experimental setup

~3 mW each LN phase modulator

BW: 100 MHz Vπ: ~ 2.3 V

PI control Loop BW: 140 kHz

Y. Ozeki et al., Opt. Express 18, 13708 (2010). Timing detection by

two-photon absorption

Synchronization performance

Two-photon signal Error signal

Time [s]

0 2 4 6 8 10

-1 0 1

Two-photon signal [V] Cavity adjustment Integrator ON

In-loop jitter: 4 fs Out-of-loop jitter: 7.8 fs

Precise (<8 fs), active, subharmonic synchronization

(BW > 100 kHz) (BW: >100 kHz)

This is the experimental setup. A Ti:sapphire laser and a Yb-ﬁber laser produced 76 MHz and 38 MHz trains of optical pulses, respectively. The pulse width is

approximately 0.3 ps. In order to achieve synchronization between the pulses, we constructed a phase locked loop. The timing difference is detected by a two-photon PD, and it is fed back to the ﬁber laser, wherein a LN phase modulator is inserted to allow high-speed control of repetition frequency. After the synchronization is

achieved, the pulses are launched to a SRS microscope. In the following slides, I’ll explain the synchronization performance and the noise level of the SRS

microscope.

This slide shows the synchronization performance of the two-color lasers. The two- photon signal increases when the pulses are overlapped in time. After the loop is closed, the two-photon signal is kept at this voltage, indicating that the lasers are synchronized. Then the integrator is turned on to achieve tight locking. The in-loop jitter and out-of-loop jitter were measured to be 4 fs and 7.8 fs, respectively. In this way, we succeeded in precise synchronization of two-color lasers.

(21)

Noise level of SRS microscope

Increase of lock-in frequency (38 MHz)

Noise level: close to the shot noise limit by 1.6 dB

Δ_{I / I}

~ 10^-6

Photo-detection circuit

Large area Si-PD (3 mmΦ) Suppression of PD saturation

High load resistance (510 Ω) Reduced thermal noise Inductor

Mitigation of CR limit

BPF & BEF avoid amplifier saturation

Then we measured the noise level of the lock-in signal of our SRS microscope. This ﬁgure shows the noise level as a function of photocurrent. The noise level increases with the photocurrent. The discrepancy between the experimental and theoretical shot noises is only 1.6 dB, which is presumably due to the signal loss in the photodetection circuit. Furthermore, the present noise level is lower than our previous system by more than 12 dB, where the lock-in frequency was 10 MHz. In this way, the shot-noise limited sensitivity is successfully achieved by increasing the lock-in frequency.

The photo-detection circuit is an important technical challenge. We use a large area silicon photodiode to suppress the saturation. To reduce the thermal noise, high load resistance of 500 Ohm is employed. To mitigate the CR limit, an inductor is inserted. Furthermore, bandpass and band elimination ﬁlters are inserted to avoid ampliﬁer saturation.

(22)

Sensitivity merit at maximum

lock-in frequency

f_m = f_rep/2 (max. freq.) 0 < fm < f_rep/2

0 fm frep

Aliasing SRS

signal

Photocurrent spectrum

Freq.

Phasor of lock-in signal Phasor of lock-in signal

0 fm = frep^/2 frep

SRS signal 6 dB

Freq.

cos sin

SRS signal Shot noise

6 dB increase

Shot noise is localized in ‘cos’

3 dB increase

Shot noise limited SNR: 3 dB, thermal noise limited SNR: 6 dB

Photocurrent spectrum

Comparison of Raman microscopy

ωS

ωCARS

Technique

Spontaneous Raman

CARS

SRS

Low

High

No

Yes

ωL ω

ω2

ω

ω2

ω1 ω ω1

High?? ^No

Finally, we point out that the maximum lock-in frequency is advantageous not only for reducing the laser noise but also for enhancing the sensitivity to a certain amount. In the photocurrent spectrum, SRS signal appears as a modulation

sideband. Simultaneously, SRS signal appears around the repetition frequency due to the aliasing effect. Therefore, if we increase the modulation frequency to the subharmonic frequency, these sidebands interfere constructively, giving 6-dB merit of SRS signal.

However, this does not directly mean the merit of signal-to-noise ratio. Actually, the shot noise increases by 3 dB at this speciﬁc frequency. Therefore the net sensitivity merit is 3 dB. This is because the lock-in reference and the pulse train are perfectly synchronized. This leads to the localization of shot noise in the in-phase

component of lock-in signal.

In this way, shot noise limited SNR is enhanced by 3 dB, and the thermal-noise- limited SNR is enhanced by 6 dB at the maximum lock-in frequency.

Since the theoretical sensitivity limit is successfully achieved, we could prove that SRS microscopy is an attractive biological imaging technique with high sensitivity and high contrast.

(23)

Comparison of Raman microscopy

ωS

ωCARS

Technique

Spontaneous Raman

CARS

SRS

Low

High

No

Yes

ωL ω

ω2

ω

ω2

ω1 ω ω1

High No

Comparison of Raman microscopy

Technique

Spontaneous Raman CARS

SRS

Low

High

No

Yes

High No

Multiplex

Multi-focus Line-scanning

Wide-field

Yes

Difficult [1]

Yes

Difficult

Possible solution: wavelength scanning

SRS is suitable for high-speed imaging with single vibrational frequency

Photodetector requires high dynamic range

[1] Ploetz et al., Appl. Phys. B 87, 389 (2007)

Nevertheless, I wouldn’t insist that SRS is the best approach. A major obstacle of SRS microscopy lies in multiplex imaging for obtaining vibrational spectrum, as well as parallel imaging such as multi-focus, line-scanning and wide-ﬁeld techniques. This is because the photodetector requires very high dynamic range so that it can receive intense laser pulses with an average power of mW. Such intense pulses may easily saturate parallel detectors. One of the solution for multiplex imaging is wavelength scanning.

At the moment, it seems like that SRS is suitable for high-speed imaging with single vibrational frequency.

(24)

Summary

• Stimulated Raman scattering microscopy

• Comparison with previous Raman

microscopy

• Theoretical sensitivity limit

• Similar to CARS

• x1000 spontaneous Raman

• Sensitivity improvement by subharmonic

synchronization technique

• 1.6 dB to the theoretical limit

Acknowledgment

• Prof. K. Fukui (Osaka Univ.)

• Assoc. Prof. S. Kajiyama (Kinki Univ.)

• Assoc. Prof. N. Nishizawa (Nagoya Univ.)

• Dr. K. Sumimura

• Mr. F. Dake, Mr. Y. Kitagawa, Mr. W. Umemura

• Mr. T. Umano, Ms. M. Ishii

To summarize, SRS microscopy employs lock-in detection to achieve label-free imaging with high sensitivity and high contrast. The theoretical sensitivity limit is similar to CARS, and 3 orders of magnitude higher than spontaneous Raman. The sensitivity limit was achieved by high-frequency lock-in detection. Our results indicate the powerful potential of SRS microscopy for label-free biological imaging with high sensitivity and high contrast.

Finally, I’d like to acknowledge the coworkers and the students. Thank you very much for your kind attention!

(25)

Thank you for your kind attention!

1Graduate School of Engineering, Osaka University

2 _JST-PRESTO

Yasuyuki OZEKI^1,2 and Kazuyoshi ITOH¹ [email protected]