Micromachined Silicon Parallel Acoustic Delay Lines as Time Delayed Ultrasound Detector Array for Real-Time Photoacoustic Tomography

Abstract

This paper reports the development of a new 16-channel parallel acoustic delay line (PADL) array for real-time photoacoustic tomography (PAT). The PADLs were directly fabricated from single-crystalline silicon substrates using deep reactive ion etching. Compared with other acoustic delay lines (e.g., optical fibers), the micromachined silicon PADLs offer higher acoustic transmission efficiency, smaller form factor, easier assembly, and mass production capability. To demonstrate its real-time photoacoustic imaging capability, the silicon PADL array was interfaced with one single-element ultrasonic transducer followed by one channel of DAQ electronics to receive 16 channels of photoacoustic signals simultaneously. A PAT image of an optically-absorbing target embedded in an optically-scattering phantom was reconstructed, which matched well with the actual size of the imaged target. Because the silicon PADL array allows a signal-to-channel reduction ratio of 16:1, it could significantly simplify the design and construction of ultrasonic receivers for real-time PAT.

1. INTRODUCTION

In biooptics, high-resolution optical imaging modalities, such as confocal microscopy [1], two-photon microscopy [2], and optical coherence microscopy [3], have played a very important role. However, their imaging resolution can be maintained only within a shallow depth (e.g., < 1 mm) due to the strong light scattering in biological tissues [4, 5]. In recent years, photoacoustic imaging has emerged as a fast growing hybrid imaging modality to break through the fundamental limitation of imaging depth. It is based on the detection of the photoacoustic (PA) waves induced by pulsed laser or electromagnetic waves. It combines the advantages of optical and ultrasonic imaging, and can provide high optical absorption contrast and deep ultrasonic penetration depth [6–8]. Based on its actual implementation [9], photoacoustic imaging can be divided into photoacoustic microscopy (PAM) and photoacoustic tomography (PAT). PAM uses a (focused) single-element ultrasound transducer to detect PA signals pixel-by-pixel with mechanical or optical scanning [10].PAT utilizes an ultrasound array system (1D or 2D) to detect multiple PA signals simultaneously without mechanical scanning. 2D or 3D PA images can be reconstructed based on a number of algorithms [11–13]. However, the construction of ultrasound receiving systems becomes more complicated and costly as the number of ultrasound transducer elements and DAQ channels increases. To address this issue, we have demonstrated a handheld probe which consisted of two arrays of parallel acoustic delay lines (PADLs) made of tightly wound fused-silica optical fiber coils [14]. Each optical-fiber PADL array can transmit eight channels of PA signals from the imaging target to a single-element ultrasonic transducer, while imposing a different acoustic time delay on each PA signal. As a result, the eight time-delayed PA signals will reach a single-element ultrasonic transducer at different times and therefore can be received unmixed. With this probe, PAT was successfully conducted by collecting and reconstructing 16 channels of PA signals using only two ultrasonic transducers and two data acquisition (DAQ) channels, which corresponds to a signal-to-channel reduction ratio of 8:1. However, although they can transmit ultrasound signals, fused-silica optical fibers are not ideal materials for constructing the PADLs. First, they are very stiff and brittle, and therefore are difficult to be tightly wound into compact coils. Second, the protective polymer jacket layer can cause acoustic attenuation and distortion especially for higher-frequency signals. Third, once they are fabricated, mechanical supporting structures are needed to hold the optical-fiber PADLs in place. However, additional acoustic loss and coupling could occur at the contacts between the optical fibers and the supporting structures. These factors limited the number of the PADLs that could be practically built and used, and therefore the signal-to-channel reduction ratio and field of view of the PAT probe.

In this paper, we report the development of a new micromachined silicon PADL array [15] and the initial demonstration of its application in real-time PAT. As an almost perfectly elastic crystal material, single-crystalline silicon is a mechanically strong material, which also has extremely low acoustic attenuation even in the MHz range [16]. Complex microstructures can be directly fabricated on a silicon substrate in one step by using photolithography and deep reactive ion etching. Therefore, single-crystalline silicon is structurally superior to optical fibers for the construction of PADLs. In this work, a 16-channel silicon PADL array was developed and its application for real-time PAT was demonstrated. The silicon PADL array was interfaced with one single-element transducer followed by one channel of DAQ electronics to receive 16 channels of PA signals simultaneously. It allowed a signal-to-channel reduction ratio of 16:1, two times higher than that of the optical fiber PADL array demonstrated previously [14, 17]. Therefore, this new acoustic time delay approach could further simplify the design and construction of ultrasonic receivers for real-time PAT.

2. DESIGN AND CONSTRUCTION

2.1. Design

Due to the high acoustic velocity (~8430 m/s [18]) of silicon, an acoustic delay length ranging from a few to tens of centimeters is necessary to provide suitable acoustic time delays (1–10 μs) between two adjacent PADLs, so that the PA signals do not overlap with each other and therefore can be clearly distinguished when they reach the single-element transducer [19]. To achieve a compact structure, the PADLs were designed as tightly wound spiral coils.

Figure 1 shows the schematic design of the longest (the 16^th) PADL. The diameter of its circular portion is 30 mm, which is still compact even though its delay length reaches 90 cm. All 16 PADLs have the same span of 60 mm between their input and output terminals (L in figure 1), which facilitates the assembly and ease of contact between the imaging target and the single-element transducer.

Figure 1.

Schematic design of the longest (16^th) silicon PADL.

Table 1 lists the main design parameters of the silicon PADL array. The shortest and longest acoustic time delays are 9.5 μs and 107 μs, respectively. The nominal incremental acoustic time delay between any two adjacent channels is 6.5 μs, which corresponds to a penetration depth of ~1 cm in soft tissue (assuming an acoustic velocity of ~1500 m/s). The cross sectional area of the silicon PADLs is 500 μm × 250 μm, which provides good mechanical stiffness (and therefore structural stability) and also good acoustic signal transmission for PA signals. For a rectangular acoustic delay line, its width or thickness (d) (whichever is smaller) should satisfy (df/V₀) << 1 to avoid signal distortion due to higher-order mode propagation and mode dispersion, where f is the frequency of the signal, and V₀ is the acoustic velocity of the delay-line material [20]. Assuming the silicon PADLs are interfaced with a single-element ultrasonic transducer with a center frequency of 2.25 MHz and a bandwidth of 100%, the maximum detection frequency will be 3.375 MHz. This value yields a df/V₀ ratio of 0.1 (<<1) to ensure single-mode transmission of the PA signals with minimal mode dispersion. Based on our previous simulation and experimental characterization results [19], the smallest radius of curvature in the center portion of the PADL structure was made larger than 2 mm to ensure high acoustic transmission efficiency by reducing mode conversion at curved sections of the acoustic delay lines.

Table 1.

Main design parameters of silicon PADLs.

PARAMETER	Value
number of channels	16
shortest delay	9.5 μs
longest delay	107 μs
incremental delay	6.5 μs
center frequency	2.25 MHz
cross section	500 μm × 250 μm
minimum radius of curvature	2 mm

Figure 2 shows a 3-D design of the assembled 16-channel silicon PADL array with one common input port and one common output port. During PA imaging, the input port is in contact with the imaging target, and all 16 channels detect the incoming PA signals simultaneously. The output port is interfaced with a single-element transducer to receive the transmitted PA signals.

Figure 2.

The assembly of 16-channel silicon PADL array (a) top view (b) side view.

2.2. Fabrication and Assembly

The silicon PADLs were fabricated on 4-inch {100}-oriented single crystal silicon wafers with a thickness of 250 μm (University Wafers, Boston, MA, USA) by using a microfabrication process developed previously [19]. First, a 300 nm-thick aluminum layer was deposited on the silicon wafers by electron-beam evaporation. Second, the deposited aluminum layer was patterned with photolithography and wet etching, forming a mask for silicon etching. Third, the silicon PADL structures were directly formed on the silicon wafers by removing unwanted regions with deep reactive ion etching (DRIE). In the last step, the etched PADL structures were separated from each other for final assembly.

To precisely position the silicon PADLs in the array at the input and output ports, two identical spacer structures were laser cut from a 0.4 mm thick clear acrylic sheet using the laser engraving and marking machine (PLS6.75, Universal Laser Systems, Scottsdale, AZ, USA) (figure 3). Each spacer structure consists of 16 shallow trenches (300 μm × 500 μm) to support the silicon PADLs and 15 interspaced deep trenches (200 μm × 1000 μm) to reduce acoustic crosstalk between adjacent supports. Both the shallow and deep trenches were formed by using a dicing saw. To protect the entire silicon PADL array, an acrylic housing was also fabricated (see figure 2). Its top panel, bottom panel, and two side walls were laser cut from a 1.5 mm thick acrylic sheet.

Figure 3.

Design for spacers of the input and output port.

The assembly process of the silicon PADL array is illustrated in figure 4. First, the bottom panel, side panels, and the two spacers were bonded together with acrylic cement, then the 16 silicon PADLs were placed in the spacer grooves. The PADLs were fastened with two small parts for mechanical stability. In the last step, the top panel was cemented in place to form a sealed housing. The input and output ends were polished to ensure uniform contact with both the imaging target and the ultrasonic transducer. Figure 5 shows the fully assembled 16-channel silicon PADL array.

Figure 4.

Assembly process of the PADL array. (a) Assembling two sidewalls, bottom, front, back, and spacers. (b) Arranging and placing 16 PADLs. (c) Fastening PADLs (d) Affixing top panel.

Figure 5.

Fully assembled 16-channel silicon PADL array. Reprinted with permission from Ref 15.

2.3. The Acoustic Acceptance Angle of Silicon PADLs

For PA imaging, the acceptance angle of the detector is an important factor to determine the spatial resolution. Based on the principle of reciprocity, the detection pattern of a detector is the same as its radiation pattern. The silicon PADLs can be treated as a linear array with a small element separation (b) (figure 6(a)). The directivity factor of its far-field radiation pattern in the lateral direction can be estimated by

$$D\left(\theta \right)=\frac{\text{sin}\left(\left(\pi d/\lambda \right)\text{sin}\theta \right)}{\left(\pi d/\lambda \right)\text{sin}\theta },$$ (1)

where d is the total width of the linear array in the lateral direction, λ is the ultrasound wavelength, and θ is the angle relative to the normal direction of the linear array [21]. Figure 6(b) shows the calculated directivity factor of the 16-channel silicon PADLs (with a total width d of 9 mm) in water at the frequency of 2.25 MHz. The radiation power quickly drops to less than −20dB when θ reaches 4°, which indicates a small acceptance angle in the lateral direction. Since all the PADLs have the same detection pattern in the elevational direction, the acceptance angle in the elevational direction the silicon PADLs can be determined with a single delay line from Eq. 1 (figure 7(a)), where d represents the width (500 μm) of the single delay line. Figure 7(b) shows the calculated directivity factor of single delay line in water at the frequency of 2.25 MHz. The radiation power only drops to −6 dB when θ reaches 55.7°, which indicates a large acceptance angle. This is mainly due to the relatively small element size compared with the ultrasound wavelength in water at 2.25 MHz. To better understand and visualize the radiation pattern, a numerical simulation was conducted to estimate the ultrasound propagation and radiation from single delay line by using Wave 2500® (CyberLogic, Inc., New York, NY, USA). Thirty-one virtual ultrasound receivers were used to measure the radiation power at different angles ranging from 0 to 90° (figure 8(a)). The directivity factor of single silicon PADL was determined based on the measured radiation power (figure 8(b)), which matches well with that calculated from Eq. 1 (figure 7(b)).

Figure 6.

(a) A linear array of PADLs (b) the directivity factor in the lateral direction calculated from Eq. 1.

Figure 7.

(a) A single PADL (b) the directivity factor in the elevation direction calculated from Eq. 1.

Figure 8.

(a) The simulated radiation pattern of single silicon delay line in Wave 2500® (b) the directivity factor estimated from the simulation data.

3. ULTRASONIC CHARACTERIZATION AND PA IMAGING

3.1. Ultrasonic Testing Setup

A two-port ultrasound testing setup was built to characterize the acoustic transmission properties of the assembled silicon PADL array (figure 9). Two 2.25 MHz transducers (V104, Olympus NDT, Waltham, MA, USA) were used to transmit and receive the ultrasound signals. The input and output ports of the silicon PADL array were placed in contact with the transmitting and receiving transducers, respectively. Mineral oil was applied between both ends of the silicon PADLs and the surfaces of the transmitting & receiving transducers to enhance the coupling efficiency and minimize unwanted reverberation. The pulser/receiver (5072PR, Olympus NDT, Waltham, MA, USA) was set to transmission mode and sent a driving voltage pulse to the transmitting transducer, which both generated ultrasound signals and also amplified signals detected by the receiving transducer. The ultrasound signals from the transmitting transducer traveled from the input port to the output port through the 16-channel silicon PADLs with distinctly different amounts of time delay. To improve the SNR (signal-to-noise ratio), all 16 channels of ultrasound signals were averaged 16 times and recorded on a digital oscilloscope (TDS2014B, Tektronix Inc., Beaverton, OR, USA).

Figure 9.

A two-port ultrasound testing setup with the 16-channel silicon PADL array. Osc: Oscilloscope. UT0, UT1:Ultrasonic transducers 0 and 1.

3.2. PA Imaging Setup

Figure 10 shows the experimental setup of the PA imaging experiment. A frequency-doubled Q-switched 532-nm Nd:YAG laser (LT-P100, Mychway Technology Co., Limited, Shenzhen, Guangdong, China) was the light source. The laser pulse duration was 8 ns and the pulse repetition rate was 5 Hz. The imaging target was a piece of silicon (1 mm × 0.25 mm × 1 mm) embedded in an optical tissue phantom (100 mm × 100 mm × 50 mm). The phantom was composed of 2% agar gelatin by weight, and the target was located 2.5 mm below the phantom surface (figure 12(a)). The PA signals generated from the imaging target propagated along the silicon PADL array and were received by a single-element 2.25 MHz ultrasonic transducer (V106, Olympus NDT, Waltham, MA, USA), then amplified by a pulser/receiver (5072PR, Olympus NDT, Waltham, MA, USA). The amplified PA signals were averaged five times and then recorded on a digital oscilloscope (TDS2014B, Tektronix Inc., Beaverton, OR, USA) at a sampling rate of 5 MHz.

Figure 10.

PA imaging setup with the 16-channel silicon PADL array.

Figure 12.

(a) Photography of an optically absorbing target embedded in an optically scattering medium. (b) PA signals received by the ultrasonic transducer through the 16 silicon PADLs. (c) Zoomed-in view of the eight signals from channel 1 to 8. (d) Zoomed-in view of the eight signals from channel 9 to 16. Reprinted with permission from Ref 15.

3.3. PA Data Acquisition and Image Reconstruction

The PA signals were averaged five times to generate one PA image; thus the imaging speed was 1 Hz. The end of each PADL on the sample surface was considered as a single element in an ultrasound array. Each sequentially detected PA signal was reshaped into 16 separate PA signals, based on the predefined time positions measured by the ultrasound transmission experiment. To reconstruct PA images, a commonly used delay-and-sum beamforming method was applied [22]. The PA images were reconstructed by summing the data from all 16 channels after compensating for the time delays in each silicon PADL, based on its length and matching phase. In a two-dimensional beam field along the x and z directions, the time delay τ_n for the n^th PADL located on (x_n, z_n) can be calculated as follows:

$${\tau }_n=\frac{\sqrt{{\left(x-x_n\right)}^2+{\left(\text{z}-{\text{z}}_n\right)}^2-R}}{c}$$ (2)

where R is the distance from the focal point to the center of the delay line, and c is the speed of sound in soft tissue, 1540 m/s. Envelope detection was applied by using Hilbert transformation along the axial direction (the z direction), followed by taking the absolute value.

4. RESULTS AND DISCUSSION

4.1. Ultrasonic Transmission through the Silicon PADL Array

Figure 11(a) shows the 16 channels of time-delayed ultrasound signals received by the single-element ultrasonic transducer. The acoustic delay in each ultrasound signal was determined by measuring the time interval between its peak intensity and that of the trigger signal (figure 11(b) and 11(c)). The average incremental acoustic delay between two adjacent channels was ~7 μs, slightly longer than designed value of 6.5 μs. This difference is due to the fact that when they travel in the curved portion of the silicon PADLs, the ultrasound signals encounter multiple reflections on the sidewalls. As a result, their velocity is different from that in a straight line and their actual travel distance is longer than the length of the silicon delay lines. The acoustic attenuation was determined based on the change in the amplitude of the ultrasound signals as they propagated through different distances. Theoretically, acoustic attenuation varies linearly with the length of the silicon PADLs, but the amplitudes of the received ultrasound signals showed otherwise. One possible reason could be non-uniform contact between the silicon PADLs and the two ultrasound transducers. Here, the average acoustic attenuation of the silicon PADLs was calculated by measuring the signal amplitude difference between the shortest (1^st) and the longest (16^th) PADL. It was 0.058 dB/cm @ 2.25 MHz, which is higher than that in a straight silicon delay line (0.015 dB/cm @ 2.25 MHz) [19]. As observed in our previous experiments [19], when ultrasound signals travel along a curved delay line, they encounter more reflections and potential mode conversions on the sidewall of the delay line, which contribute to higher acoustic attenuation. However, it should be mentioned that the attenuation of 0.058dB/cm is still very low, and can be neglected.

Figure 11.

2.25 MHz ultrasound signals propagating through the 16 silicon PADLs. (a) A trigger and 16 signals. (b) Zoomed-in view of a trigger and the eight signals from channels 1 to 8. (c) Zoomed-in view of the eight signals from channel 9 to 16. Reprinted with permission from Ref 15.

4.2. PAT Imaging

Figure 12(b), 12(c), and 12(d) show the 16 PA signals from the silicon PADL array acquired by the single-element ultrasound transducer and the DAQ channel after five times averaging. Each PA signal was identified based on the time delay (7 μs) previously determined in ultrasound transmission testing. Figure 13 shows the reconstructed PA image in Matlab®. The envelope of the PA signals was detected by using Hilbert transformation, and the PA image was reconstructed using the synthetic aperture focusing technique (SAFT) [23] to improve spatial resolution. As shown in figure 13(a), in the reconstructed PA image, both the location and the size of the imaging target match the actual values. The image contrast, defined as (PA_target − PA_background)/ PA_background, is calculated to be ~1.9. The spatial resolution, defined as the one-way distance between 10% and 90% of the maximum, divided by the minimum, is ~2.1 mm. Figure 13(b) shows the enhanced PA image after applying a signal threshold of 30%.

Figure 13.

(a) Reconstructed PA image of the imaging target. (b) Thresholded (30%) PA image of (a). Reprinted with permission from Ref 15.

5. CONCLUSION

In this work, we have successfully demonstrated a micromachined silicon PADL array for real-time PAT. The array capitalizes upon the extremely low acoustic loss of single-crystalline silicon and the high precision of the micromachining process. Using the silicon PADL array, a transducer/channel reduction ratio of 16:1 was achieved, which is twice what was obtained previously with optical-fiber PADLs. In the future, larger silicon PADL arrays will be investigated to provide even higher transducer/channel reduction ratios to fulfill the potential of the time-delayed ultrasound reception approach. In addition, new solutions will be developed to improve the sensitivity, the acoustic acceptance angle in the elevation direction, and impedance matching at the input terminals to obtain better PA signal detection.