A sensitive, low noise, DC to 12 MHz, large area photodiode preamplifier for photothermal heterodyne imaging

Photothermal heterodyne imaging (PHI) is a technique that has been used for numerous applications and is seeing increased utility for spectroscopy with the development of tunable laser sources. In PHI, a species (particle or molecule) is detected when energy is absorbed from the pump beam and is transferred as heat to the surrounding medium. This heat transfer creates a nano-lens that affects the propagation of the probe beam and changes the intensity of the probe at the detector.^1,2 The advantages of PHI are that the signal is free from background and is not strongly affected by scattering, as the signal only arises from the absorbing species that dissipates heat. Most PHI measurements rely on lock-in amplifiers for detection, as the change in refractive index with the temperature (∂n/∂T) of the surrounding medium is quite small (e.g., 10⁻⁴ K⁻¹ for a typical organic solvent). Several developments have been proposed to improve the signal to noise ratio,^3,4 such as the phase transition method,^5–7 and the detection of single molecules at room temperature has been demonstrated.⁸ 

In order to attain high sensitivity, a low-noise photodiode preamplifier is needed. While high-end commercial products are available, it is straightforward to design an inexpensive, high performance preamplifier, yet reports are lacking in the literature. Our initial PHI experiments⁹ were noise limited, where the system’s noise level was observed to be 1-10 μV. This noise level inhibited applications that required low excitation pump/probe power to minimize the temperature increase in a sample. Briefly, our first PHI setup was based on a 10 kHz mechanical chopper, Stanford SR530 lock-in amplifier, and Thorlabs PDA100A Si photodiode amplifier (100 mm²). In this initial setup, the noise primarily arises from two sources: a slow modulation frequency and the noise of the photodiode. First, the 10 kHz modulation speed is too slow. Laser noise at low frequencies is larger due to the 1/f noise contributions and low frequency thermal and mechanical fluctuations. These noise effects all decrease at a high frequency (e.g., >1 MHz).¹⁰ Second, while inexpensive, the noise from the large area PDA100A detector is also relatively high. It is well known that the capacitance of the large area photodiode is usually quite large (e.g., 2000 pF), and this capacitance has a significant effect on the noise performance of the circuit due to the noise gain.¹¹ To achieve low noise, a preamplifier is typically designed for the large area photodiode; however, this generally increases the cost of the photodiode detector and is not true for the PDA100A. An additional complication is the T-feedback design¹² that also decreases the performance when used at higher gain.

In PHI, low noise and a fast photodiode preamplifier are quite important for high-sensitivity and high-speed imaging in which a small lock-in time constant is used. In general, the smaller the time constant, the larger the noise. In this report, we present a simple transimpedance photodiode preamplifier, which was built using an ADA4899-1 op-amp as the low noise input stage. This photodiode preamplifier has a bandwidth from DC to 12 MHz and 7 nV/Hz^1/2 noise level, which is sufficient for most PHI experiments. This bandwidth encompasses the 100 KHz–MHz modulation frequencies typically used by others for PHI experiments.

PHI is a two-color nonlinear spectroscopic technique where the magnitude of the signal depends on the power of both the pump and the probe beams. The probe beam wavelength is selected such that it is not absorbed by the sample, and therefore the probe beam intensity is designed to respond only to refractive index changes. Because the probe beam is not absorbed, very high powers can be used, and the maximum probe power applicable is ultimately limited by the sample damage threshold. This higher power probe beam intensity for PHI is different than other high sensitivity detection methods, which typically detect low light or current (e.g., tunneling current in STM^13,14). The power of the probe beam incident on the photodiode detector is usually above 1 mW, sometimes 10 mW or more. A 1 mW probe beam incident on a photodiode with 0.5 A/W sensitivity will generate a current of 0.5 mA, which corresponds to a noise level of 12.6 nV/Hz^1/2 using a voltage gain resistor of 1 kΩ. If the total input noise of the photodiode preamplifier is below 12.6 nV/Hz^1/2, the system will approach shot-noise limited performance for probe beam powers above 1 mW incident onto the photodiode detector.

Previous reports have utilized a low noise photodiode preamplifier based on an LC resonant amplifier.^15,16 However, the LC resonant amplifier has some limitations: First, it is more complicated as it uses discrete transistors and requires a high voltage supply. Second, the modulation frequency is determined by the resonant frequency and is usually fixed. However, it is advantageous to have a tunable modulation frequency for PHI measurements. It was previously reported that the PHI signal decreases when the modulation frequency increases.^2,3 To maintain a high signal, low noise, and a reasonably fast imaging speed, it is important to optimize the modulation frequency.^17,18 Third, the LC resonant amplifier also requires a stable external modulation frequency matched to its resonance frequency. Fourth, the Q factor of LC is not easily tuned. Tuning requires adjusting the values of L or C, but the values of L and C also set the resonance frequency. Additionally, the Q will affect the bandwidth (or time constant), which will conflict with the time constant setting in the lock-in amplifier.

To simplify the design, our preamplifier is based on a straightforward and convenient op-amp transimpedance configuration. The circuit diagram is shown in Fig. 1. Here we use a large area Hamamatsu S3071 photodiode (diameter = 5 mm). The datasheet specifies that the capacitance of S3071 is 70 pF at 1 V bias and reduces to a nearly constant 18 pF when the bias increases above 10 V. As an input stage, we used the ultra-low noise ADA4899-1 op-amp, U₁, with the voltage gain resistor (R_f) of 1 kΩ. The ADA4899-1 is an ultralow noise (1 nV/Hz^1/2) and distortion (<−117 dBc at 1 MHz) unity-gain stable voltage feedback op-amp, with a wide supply voltage ranging from 4.5 V to 12 V, a low offset voltage (35 μV), a low input bias current (100 nA), a slew rate of 310 V/μs, and a wide bandwidth (600 MHz). Our design goal is a noise level of 10 nV/Hz^1/2 and a bandwidth of 10 MHz, so we use a 12 V photodiode bias to reduce its capacitance and increase its response. R₁ and C₁ form a low pass filter to decrease the possible high frequency noise that can be transferred from the power supply. The value of R₁ should not be too high, as it will decrease the effective bias when a high current passes though the photodiode. C₂–C₆ are power supply bypass capacitors, and R₂ sets the output impedance and also limits the output current. The compensation capacitor C_f is used to control the frequency response. In our system, we can easily change the voltage gain by switching R_f (matched C_f) from 50 Ω, 300 Ω, 1 kΩ, and 3 kΩ which depends on the power of the probe beam. Our design is simple, highly sensitive, and easy to use. We can choose any modulation frequency between DC and 12 MHz. The gain resistor could be easily switched to match the probe beam’s power from below 1 mW to several tens of mW. Our noise level is well below that of a commercial lock-in amplifier, providing nearly shot noise limited performance for PHI applications.

LTspice software was used to simulate the frequency response and noise characteristics of the photodiode preamplifier. In the simulation, we used 20 pF for the capacitance of photodiode, and the results are shown in Fig. 1. The frequency response Fig. 2(a) shows that at the −3 dB (0.707) cut-off point the bandwidth is around 35 MHz, and its frequency response is flat from DC to 10 MHz range. The noise characterization curve in Fig. 3(b) shows that the total output noise is about 7 nV/Hz^1/2 at 10 MHz frequency. To test the real frequency response, we set the frequency value on the SR844 lock-in amplifier and sent this reference signal to drive the acousto-optic modulator (AOM) to switch the on/off state of the 633 nm probe beam, which illuminates the photodiode. The output of the preamplifier is sent to a Tektronix TDS2024B (200 MHz, 2 GS/s) oscilloscope through a 2-3 m long BNC cable. The 1 MHz and 5 MHz frequency responses are shown in Figs. 3(c) and 3(d); we used the pulse rise time to calculate the bandwidth of the preamplifier based on the following equation:¹⁹ 

where t_rise is 28 ns in the 5 MHz test and the calculated bandwidth is 12.5 MHz, which is a little lower than the simulated 35 MHz. This discrepancy appears to arise from R₂ in combination with the capacitance of the BNC cable forming a low pass filter, which limits the output response of the ADA4899-1. Usually the capacitance of the RG58 BNC cable is about 100 pF/m, equating to a 200-300 pF capacitance for the 2-3 m long BNC cable used, which combined with a 50 Ω resistor will form a 10.6 MHz–15.9 MHz low pass filter. However, the observed 10 MHz bandwidth achieves the design goal for our PHI experiments.

To characterize the noise, we used the signal from the SR844 lock-in amplifier to evaluate the preamplifier’s noise level. The noise characteristics are shown in Table I. The lock-in amplifier has an intrinsic 10 nV output noise at 1 MHz. When connected to the preamplifier with a 1 mW 633 nm laser incident on the photodiode, the noise level does not increase. This indicates that our preamplifier’s noise level is below or close to the noise level of the SR844 (8 nV/Hz^1/2 maximum input noise) and also indicates that the light output from the He—Ne laser is quite stable. However, the noise increases when we change to either a 785 nm laser or a 532 nm laser. For the 785 nm laser, the noise increase is small, but for the 532 nm laser, the observed noise is quite large. This suggests that our 532 nm laser’s output is less stable and contains some high frequency oscillations. Further evidence that the 532 nm is noisier than the 633 nm laser was the increased output noise observed on the AC input of the oscilloscope.

To further characterize the performance of our preamplifier, the relative laser noise as a function of the sampling frequency and the noise dependence on optical power were measured, as shown in Fig. 3. At a low frequency, the relative noise of the laser behaves like 1/f noise; see Fig. 3(a). The CW He—Ne laser reached shot noise performance at a frequency of about 300 kHz. This frequency is lower than that typically needed with picosecond or femtosecond pulsed lasers, which usually require up to several MHz (e.g., 5 MHz). The power noise arising from our preamplifier showed a square root dependence (slope = 0.5 in log scale) on the power from 0.1 to 4 mW, see Fig. 3(b), which indicates that shot noise limited detection is obtained at powers above 1 mW.

We benchmarked the performance of our preamplifier design against a Thorlabs PDA100A. The noise characteristics are shown in Table II. While the gain settings are different, we can compare results in terms of expected noise limits. The thermal noise of the gain resistor scales as √R, where R is the value of resistor. For 1 kΩ and 1.5 kΩ resistors, the expected noise is about 4 nV/√Hz and 4.8 nV/√Hz at room temperature, respectively. The noise behavior of our design is much closer to the thermal noise limit than the PDA100A. Additionally, our design shows a smaller noise increase upon switching to a relatively higher gain setting.

The experimental PHI setup is based on a home-built transmission mode microscope equipped with an Olympus 50× air objective (NA = 0.5, LMPLFLN50xBD) for excitation and a Nikon 50× air objective (NA = 0.45, WD = 17 mm) for collection, as diagramed in Fig. 4.

The pump beam consists of a 532 nm laser (Innovation Photonic Solution, I0532sl0100MF). The pump laser passes through an acousto-optic modulator (AOM) (Gooch and Housego, AOMO 3200-124) which typically is modulated at 400 kHz. The 633 nm probe beam is provided by a He—Ne laser (Melles Griot, 25-LHP-925). Both beams are focused onto the sample, and careful alignment is needed to make sure that the focal spots have good overlap; this is achieved by using achromatic microscope objectives and by using pairs of collimation lens for fine adjustment. Here we used a Nanonics-MV4000 piezo stage scanner for sample scanning. A bandpass filter (Semrock, LL01-633-12.5) was used to block the pump beam and pass only the probe beam to focus onto the photodiode. The generated photo-current signal is amplified by a transimpedance amplifier first and then sent to the lock-in amplifier (Stanford Research Systems, SR-844) to demodulate the quite weak photothermal signal.

We demonstrated the performance of our setup with photothermal microscopy on 80 nm diameter gold nanoparticles (NPs). The gold NPs (80 nm citrate NanoXact gold) were purchased from nanoComposix (San Diego, CA). The Au sol was centrifuged twice before the optical measurements to remove excess reactants and surfactants. A small drop of the sol was placed on a clean cover glass substrate. The true color dark field image is shown in Fig. 5(a). Isolated single nanoparticles usually show up green; however, bright yellow indicates aggregated nanoparticles or other particulates. Our sample shows a distribution of particle sizes, and small individual nanoparticles are evident as dimmer green spots. These dim scatterers are also detected in the PHI image shown in Figs. 5(b) and 5(c). The small NPs also evidence a weak PHI signal. Figure 5(d) shows the cross section along the white line in Fig. 5(b). As indicated earlier, the noise in the PHI system is measured to be 10 nV, and the signal to noise ratio (SNR) is about 100 for a 1 μV PHI signal in air. With our current configuration (0.5 NA, 0.3 mW pump, and 1 mW probe, test in air), calculations indicate that the sensitivity is sufficient to detect a 20 nm Au NP. The signal of a 20 nm NP will be 1/64 of an 80 nm NP, based on the change in volume; however, the SNR can be easily improved by increasing the pump/probe beam’s power density (e.g., using a high NA objective or higher excitation power). As the power increases, the tradeoffs are photon damage and heating effects. The temperature will further increase if the sample absorbs at the probe wavelength. Additionally, changing the medium from air to water will further increase the PHI signal by 10×.⁹ Incorporating our preamplifier with a high NA objective (NA = 1.4) to increase the power density and using a glycerol medium should enable the imaging of small (e.g., 5 nm) Au NPs.

A simple low noise, large area photodiode preamplifier is demonstrated. Its 7 nV/Hz^1/2 noise level and DC to 12 MHz bandwidth is enough for most PHI experiments. Due to its low noise level, high bandwidth but also large detection area, this design will be useful for other pump-probe detection techniques, such as stimulated Raman scattering (SRS).²⁰ 

This work has been supported by the National Science Foundation, Award No. CHE-202655.