1. Introduction

Silicon photonics integrated circuits (Si PICs) are now at the heart of high-speed, energy-efficient fiber-optical communication transceivers utilized within data centers [1,2]. Si PICs are major enablers of distributed processing of large-scale AI/ML models [3,4]. In these scenarios, data processing still predominantly takes place in the electrical domain, necessitating opto-electric (O/E) and electro-optic (E/O) data domain conversions. Our hypothesis is that performing appropriate signal processing on the optically-modulated information signal may reduce the power consumption and latency across the system while bolstering security. Optical signal processing may occur at different remote nodes across an optical network [5,6]. Providing uninterrupted and reliable electrical power and control signals to some of the nodes may be costly or not possible. In these cases, the required power and control signals may be provided optically from a remote site. A remotely powered, controlled, and monitored optical switching, radio over fiber system, and data sequence detection was demonstrated in a benchtop setting [7–10]. It is possible that other systems such as optical phased array lidars and radio over fiber transceivers implemented in Si PIC [11,12], where the information already resides in the optical domain, benefit from optical signal processing as well.

In this paper, we demonstrate photonic integrated circuits (PIC) capable of performing basic optical signal processing tasks: (1) amplitude modulation and switching, (2) two-tap sequence detection. The proposed all-optical remotely controlled and biased signal processing system uses optical signals to bias and control the optical processing unit. Different wavelengths may be used to carry the data, control, and bias optical channels. The control and bias signals were delivered in the same fiber link with the main optical data in a WDM configuration. The conceptual schematic of our all-optical signal processor is shown in Fig. 1. A wavelength demultiplexer, realized as an array of series-coupled microring resonators, is used to separate the optical signals that carry data, control, and bias. The optical data provides the input to the optical signal processor. The control and biased optical signals are fed to optical-power-to-voltage converters realized as series stacked photodiodes operating in photovoltaic mode. The generated voltages are then used to bias and control the optical signal processor.

 

Fig. 1. Conceptual diagram of the all-optical signal processor.

Download Full Size | PDF

This work is realized in the Tower Semiconductor commercial silicon photonics process (PH18 MA) featuring silicon thickness of 220 nm on top of a 3 µm buried silicon oxide layer. Single mode strip Si waveguide with 500 nm width is used in most of the designs.

2. Building blocks

The complexity of the optical signal processing unit determines the wavelength demultiplexer number of channels and the optical voltage converter maximum output voltage. In this section, different design sizes are introduced to accommodate different processing unit complexities.

2.1 Wavelength demultiplexer

The wavelength demultiplexer, shown in Fig. 2(a), is implemented using an array of microring resonator filters. To be more tolerant to the fabrication and temperature variations, the filters are designed to have a wideband and flat-top response. We used five identical series-coupled ring resonators to implement each filter [13,14]. The coupled ring resonators have a race-track geometry with low loss 180° Euler bends. The resonators were designed to have a small strip waveguide width of 0.4 µm so that the required coupling coefficient can be achieved with a smaller coupling length. The Euler bends have end-to-end gap of 5 µm and simulated loss of 0.07 dB. These result in a high-quality factor (Q) resonator with a smaller roundtrip length (L_rt) and wider free spectral range (FSR). Using symmetric coupling coefficient between the coupled rings introduces large ripples in the passband region of the filter. To achieve a flat top response from the series coupled ring resonator, their power coupling coefficient from top to bottom (including bus-ring coupling as shown in Fig. 2(a), are designed to be 0.45, 0.09, 0.05, 0.05, 0.09, and 0.45, respectively [14]. The coupling coefficients are controlled by adjusting the corresponding coupling gaps. The gaps between the bus-ring and the ring-ring waveguides from top to bottom are 0.2 µm, 0.32 µm, 0.37 µm, 0.37 µm, 0.32 µm, and 0.2 µm, respectively. Three different wavelength demultiplexer designs were fabricated with number of channels (N_ch) of 3, 5, and 7. Figure 2(b) is a micrograph of a 3-channel wavelength multiplexer. The roundtrip lengths (L_rt) of the first channel resonators’ are 25.7 µm. The L_rt of the subsequent filter channels are incremented by 0.24 µm, 0.14 µm, and 0.11 µm steps for the three, five, and seven channel designs, respectively. These values make the channels equally spaced within the FSR of the ring resonators. The measured transfer function of the drop ports of the ring resonators are shown in Fig. 2(c). In all the measurements in this paper, grating couplers and fiber array are used to couple the optical signals into and out of the fabricated chip. The total measured coupling loss is around 8 dB, this includes both the grating coupler and the fiber array losses. The insertion loss of the filters varies between 2 to 5 dB. Each filter has 3 dB bandwidth (BW_-3dB) of 2.4 nm and free spectral range of 23 nm. The measured filter responses aren’t perfectly flat top likely due to the fabrication mismatches between the five coupled rings round-trip phases and the coupling between them. Behavior model Monte-Carlo simulation was used to check the yield of the filter response with random refractive index and coupling coefficient variations; the transfer function achieved acceptable ripples (<2 dB) for 95% of the samples. This fabrication variation can be later calibrated using micro-heaters on top of the coupled rings and the coupling regions. The channels center to center spacing are 7.6 nm, 4.6 nm, and 3.3 nm for the three, five, and seven channel designs, respectively.

 

Fig. 2. Wavelength demultiplexer fabricated designs and measurement results. (a) Schematic diagram of the series coupled ring resonator-based wavelength demultiplexer showing the main design specs. (b) Micrograph of the fabricated 3-channel wavelength demultiplexer. (c) Measured transfer function of three different wavelength demultiplexer designs with different number of channels (N_ch), 3-channel, 5-channel, and 7-channel.

Download Full Size | PDF

2.2 Optical power to voltage converter

A set of series stacked photodiodes operating in the photovoltaic mode (Fig. 3(a)) are used to convert the optical control and bias signals to the corresponding DC control voltages depend on the optical power values of these signals [7,15]. The maximum voltage generated across one non-biased diode operated in the photovoltaic mode is limited by the photodiode open circuit voltage (≈0.35 V for Germanium photodiode). By connecting the photodiodes in series configuration, the maximum generated voltage can be expanded as the voltages across each diode are added in series. A multi-mode interferometer (MMI) based distribution network is used to deliver the input optical control signal to the photodiodes. The photodiode in the photovoltaic mode has the advantages of not needing electrical bias; but, on the other hand, it is inefficient for driving a low resistance load (it works best with capacitive loads that do not drain DC current). Hence, it is more suitable to bias charge depletion modulators and not charge injection or thermo-optic modulators. Five different optical power to voltage converters with numbers of series photodiodes (N_pd) of 1, 2, 4, 8, and 16 were fabricated. We used the PH18 MA Tower Semiconductor PDK Germanium on Silicon vertical P-I-N photodiode, where a germanium layer is formed on top of the Si strip waveguide to absorb the light. The photodiode width is 8.6 µm and the length is 15 µm. Tapers are used to connect the 0.5 µm MMI network waveguides to the 8.6 µm PD waveguides. A micrograph of the fabricated optical to voltage converter with N_pd= 16 is shown in Fig. 3(b). Figure 3(c), shows the current-voltage characteristic of the fabricated photodiode when the input power ≈ 3 mW. The measured open circuit (I_D= 0) voltages of the combined series photodiodes are 0.35 V, 0.7 V, 1.25 V, 2.35 V, and 4.2 V, respectively. The measured converted output DC voltage versus the input optical power is shown in Fig. 3(d). As expected, the output voltage starts to saturate as it reaches the diode open circuit voltages. The optical power to voltage converter design with N_pd= 16 converts a 1.5 mW input optical power to 3.75 V output DC voltage. The control signal speed is limited by the optical to voltage converter bandwidth. In our design using photovoltaic diodes, the measured bandwidth is around 10 MHz which is sufficient for controlling or calibrating the optical processor. On the other hand, using photoconductive photodiodes can extend the bandwidth to GHz range, but it requires biasing.

 

Fig. 3. Optical power to voltage converter fabricated designs and measurement results. (a) Schematic diagram of the series stacked photovoltaic photodiodes. (b) Micrograph of the fabricated optical to voltage converter with 16 stacked photodiodes, with a zoomed-in micrograph of 4 photodiodes. (c) Measured current voltage characteristic of the stacked Si-Ge photodiodes for different numbers of series photodiodes: 1, 2, 4, 8, and 16. (d) Measured open-circuit output voltage versus input optical power for different numbers of series photodiodes (N_pd): 1, 2, 4, 8, and 16.

Download Full Size | PDF

3. All-optical signal processing demonstration

In this section, we demonstrate two examples of the all-optically controlled and biased signal processor using the building blocks mentioned in the previous section. The first prototype is an amplitude modulator, and the second prototype is a two-tap sequence detector.

3.1 Amplitude modulator

One of the simplest optical signal processors is the amplitude modulator which is used in most of the integrated photonics systems. Our amplitude modulator is implemented with balanced Mach-Zehnder Interferometers (MZI) with charge depletion phase modulators, as shown in Fig. 4(a). The driving voltages of these phase modulators are derived optical power to voltage converters. The required driving voltage of these modulators should be small; otherwise, the required optical power needed to generate the driving voltages will be large. Thus, these modulators should be designed to be long and efficient (low V_π.L). The length of the implemented modulators is 4.85 mm limited by the 5 mm × 5 mm available die area. The ridge waveguide used to implement the phase modulator has a strip width of 0.5 µm. To improve the modulator efficiency, a shifted pn-junction modulator geometry is used [16,17]. It is well known from the plasma effect that the hole concentration change introduces more phase shift to the optical field than the electron concentration change [18]. Therefore, the pn junction is shifted by 50 nm from the center to maximize the hole density in the center of the waveguide where the optical field is maximum. A standalone MZI is used to test the 4.85 mm length charge depletion modulator; the measured V_π is around 3.75 V (V_π.L_π=1.8 V.cm). The amount of required optical power at the optical to voltage converter input to generate that V_π voltage is 1.5 mW. The amplitude modulator is a balanced design with phase modulators in both arms of the MZI to achieve a high extinction ratio. The control signals feeding the modulators are either CW or low-speed modulated waveforms. Therefore, the modulators need not be designed to support high modulation speeds. The two phase modulators are controlled independently through separate optical power to voltage converters each realized as 16 series stacked photodetectors. Three wavelengths, one for input data and two for phase modulator control voltages, are needed in the proposed WDM scheme. A three-channel wavelength demultiplexer is used to split these signals. The wavelengths used for the data and control signals are centered at 1546 nm, 1553 nm, and 1559 nm, respectively. The total active area of the fabricated all optical amplitude modulator is around 0.6 mm², as shown in Fig. 4(b). To measure the all-optical controlled performance of the amplitude modulator, two tunable laser sources (one for the input power and one for control signal) are combined using a 50:50 coupler. The output of coupler carries the input wavelength division multiplexing signal that is then inserted to the fabricated chip. The measured output optical power versus the optical power of the control signals at each of the phase modulators in the balanced MZI are shown in Fig. 4(c). Due to fabrication mismatches between the two MZI arms, the response of the amplitude modulator as a function of the two phase modulators are not identical. The measured results show that the amplitude modulator can be fully tuned using 0.25 mW and 18.8 mW optical power when controlling PM₁ and PM₂, respectively. The corresponding output power peak to null ratios are 15 dB, and 18 dB, respectively. By using a 2 × 2 coupler at the output of the balanced MZI instead of the MMI, an all-optically controlled optical switch may be realized.

 

Fig. 4. Optically-controlled amplitude modulator: (a) schematic diagram, (b) fabricated device micrograph (c) measurements showing the output power versus the input optical control powers that control the two arms of the amplitude modulator MZI.

Download Full Size | PDF

3.2 Two-tap sequence detector

To demonstrate the proposed all-optical controlled signal processing system with a more complex processing unit, an all-optically-controlled two-tap symbol sequence detector is implemented. This function can be obtained by coherently adding the input data stream to a delayed version of itself. This can be implemented using two feedforward waveguides in an asymmetric MZI scheme [10,19] where one branch includes a delay line equal to the symbol rate of the modulated optical input data. The functional operation is as follows: (1) the input optical data stream is equally divided to two branches; (2) the data in one branch is delayed by a symbol period; (3) the carrier phases of both branches are matched, and (4) the resulting signals from two branches are combined again.

As an illustrative example, let us assume an OOK modulation where the optical power is ${P_H}$ for the “1” symbol and ${P_L}$ for the “0” symbol. Assuming lossless implementation, the optical output power level will be ${P_H}$ when two consecutive ones (11) appear at the input, ${P_L}$ when two consecutive zeros (00) appear at the input, and $0.25({\surd {P_L} + \surd {P_H}} )^2$ when a sequence of (01) or (10) appears in the input. Therefore, this scheme can discriminate (00), (11), and (10)/(01) sequences. To discriminate (10) versus (01), intentional mismatch can be introduced in the loss of the two arms of the asymmetric MZI. As an example, assume the optical loss in the delayed arm in the otherwise lossless structure is $L < 1$. The output power levels corresponding to (00), (01), (10), and (11) sequences of an OOK modulation will be ${P_{00}} = 0.25{P_L}({1 + \surd L} )^2$, ${P_{01}} = 0.25({\surd {P_L} + \surd L\surd {P_H}} )^2$, and ${P_{10}} = 0.25({\surd {P_L} + \surd L\surd {P_H}} )^2$, and ${P_{11}} = 0.25{P_H}({1 + \surd L} )^2$, respectively. These are four distinct levels that can discriminate the input consecutive symbol sequences, however these levels cannot be equally spaced. In a special case where ${P_L} \cong 0$ and $L \cong 0.5$, the output optical power levels corresponding to (00), (01), (10), and (11) sequences will be $0$, $0.125{P_H}$, $0.25{P_H}$, and $0.73{P_H}$, respectively. Figure 5 shows the modelled output of a two-tap sequence detector in this case.

 

Fig. 5. Conceptual diagram of the two-tap sequence detector showing the dynamic response for an OOK input data.

Download Full Size | PDF

The schematic level diagram of the all-optically controlled two-tap sequence detector is shown in Fig. 6(a). In this design, the delay line is constructed in a meandered layout consisting of low-loss multi-mode straight waveguides, single-mode Euler waveguide bends, with parabolic tapers connecting the two [20]. An amplitude modulator is realized before the asymmetric MZI to control the input data power ratio in the MZI arms. This ratio may be adjusted to provide the desired discrimination levels in the detected consecutive bit sequences. The amplitude modulator implementation is similar as that described in the previous section, a balanced MZI with charge depletion phase modulators. A 2 × 2 coupler is used at the output to route the weighted portion of the optical data to the two arms of the asymmetric MZI. Charge depletion phased modulators are used in the asymmetric MZI arms to compensate for the phase mismatch due to the fabrication mismatch and the delay line induced phase. This system uses a three channel wavelength demultiplexer to separate the input data steam and the two optical signals that control the amplitude and phase modulators. The two control optical signals are then converted to electrical voltage using an optical to voltage converter with 16 series stacked photodiodes. To reduce the design complexity, only the phase modulator in one arm of the amplitude modulator and the asymmetric MZI are controlled. The fabricated design micrograph is shown in Fig. 6(b). While a two-tap scheme is demonstrated, the approach may be extended to larger number of taps for detecting longer data sequences. The number of taps can be extended by using lattice configuration (cascaded MZI) with incremental delay values equal to the integral multiplication of the symbol rate. For example, a 4-tap correlator will have two cascaded MZIs with delay lines of 0 T, 1 T, 2 T, and 3 T [10].

 

Fig. 6. Optically-controlled two-tap sequence detector: (a) schematic diagram, (b) fabricated device micrograph (c) VNA frequency sweep measurements showing the optical transmission for different values of the optical control signal that feeds the amplitude modulator.

Download Full Size | PDF

The first step of the measurements is to calibrate the amplitude modulator and the asymmetric MZI phase difference. The amplitude modulator can be calibrated by observing the extinction ratio of the system output as the input wavelength or frequency is continuously swept. We used a vector network analyzer (VNA) to perform this measurement. Port 1 of the VNA goes through an external optical modulator to modulate a laser signal. The modulated signal is then fed to the chip. The output is connected to port 2 of the VNA. Measured S₂₁ responses for different values of optical control power signal (P_λc1ip) are shown in Fig. 6(c). The extinction ratio (peak to null ratio) is maximum when the amplitudes of the two MZI arms (E_A and E_B points) are balanced. The results show that the amplitude modulator control signal power (P_λc1ip) of 2.5 mW, corresponding to the maximum extinction ratio, makes |E_A|=|E_B|. When P_λc1ip is less than 2.5 mW the optical data power at the delayed (upper) arm is less than the reference (bottom) arm (|E_A|<|E_B|), and vice versa when P_λc1ip is more than 2.5 mW, |E_A|<|E_B|. The phase difference between the two arms of the asymmetric MZI can be calibrated by observing the peak and the null point at the output while changing the phase modulator optical control signal (P_λc2ip). When the optical signal at the end of the asymmetric MZI arms are in phase (out of phase), the output is maximum (minimum). The P_λc2ip value of 0 mW and 4.2 mW corresponds to these two cases, when the amplitude modulator is in the equal power configuration (|E_A|=|E_B|).

In order to demonstrate the dynamic measurements, we used an arbitrary waveform generator to generate PRBS and external modulator to generate the optical data to the correlator. Along with the optical data, two optical signals carrying the control signals are combined in a WDM configuration and coupled to the chip. The PRBS optical data is on channel 2 centered at 1546 nm. The optical control signals are inserted on channel 1 (controlling the phase difference of the asymmetric MZI) and channel 3 (controlling the amplitude modulator) centered at 1559 nm and 1553 nm, respectively. The output is then detected using a digital storage oscilloscope. The PRBS data rate is 1.25 Gbps as the delay line used in the two-tap sequence detector has delay of 0.8 ns.

Figure 7(a) shows the measured output when the optical data at the asymmetric MZI arms’ ends are in phase (P_λc2ip ≈ 0 mW) – this implies that two consecutive bits are added coherently at the output. The optical output is highest when the two consecutive bits are 11, lowest in case of 00, and midway for 10 and 01 cases. The results are reported for three cases. In a first case, P_λc1ip = 2.5 mW resulting in |E_A|=|E_B|; in this case, the (10) and (01) sequences result in the same output power levels. In a second case, P_λc1ip = 1 mW resulting in |E_A|<|E_B|; in this case, the later optical bit has larger optical power than the earlier (delayed) optical bit, hence the (01) sequence has larger optical output than the (10) sequence. Finally, in a third case, P_λc1ip = 5 mW resulting in |E_A|>|E_B|; in this case, the (10) sequence has larger optical output than the (01) sequence. The second and the third cases represent the desired sequence detection configurations as all the four sequences are discriminated.

 

Fig. 7. Optically-controlled two-tap sequence detector dynamic measurements for OOK input modulated signal. PRBS OOK signal is the input to the correlator (in Red), the optical output of the modulator (in Blue) for three different cases of optical power ratios at the end of the asymmetric MZI branches (three amplitude modulator settings). Top: the power ratio is the same (|E_A|=|E_B|), middle: the power in the non-delayed arm is larger (|E_A|<E_B|), and bottom: the power in the delayed arm is larger (|E_A|>E_B|). This is shown for two cases when the phase difference in the optical fields at the point Ea and EB are in phase and out of phase, (a) and (b) respectively.

Download Full Size | PDF

Similar measurements may be conducted (Fig. 7(b)) for the case where the optical data at the asymmetric MZI arms’ ends are out of phase (P_λc2ip ≈ 4.2 mW). In this scenario, each two consecutive bits are subtracted, hence the (11) and (00) result in low output power levels, while the (10) and (01) sequences result in high output power levels. The results are reported for the same three amplitude modulator cases and optical control signal (P_λc1ip) values described in Fig. 7(a).

We also tested the two-tap sequence detector with higher order modulations, PAM-3 and PAM-4 with the same symbol rate of 1.25 Gbaud (Fig. 8). As there are more combinations of two-bit sequences, for clarity the results are only shown for one specific case: the optical data at the asymmetric MZI arms’ ends are in phase (P_λc2ip ≈ 0 mW) and have the same power (P_λc1ip = 2.5 mW (|EA|=|EB|)). In this case, there are 6 output optical levels for the PAM-3 input and 10 levels for the PAM-4 input.

 

Fig. 8. Optically-controlled two-tap sequence detector dynamic measurements for PAM-3 and PAM-4 input modulated signals in (a) and (b) respectively. PRBS signal is the input to the correlator (in Red), the output of the modulator (in Blue) for the same optical power ratios at the end of the asymmetric MZI branches (one amplitude modulator setting), the power ratio is the same (|E_A|=|E_B|). These results are when the optical fields at the point E_A and E_B are in phase.

Download Full Size | PDF

4. Conclusion and summary

We have demonstrated an integrated optically-controlled amplitude modulator and two-tap sequence detector that utilizes a WDM scheme to separate the data and control channels. Wideband coupled ring resonator filters are used to separate the control and data channels (BW_-3dB = 2.7 nm & FSR = 23 nm). The optical signals from control channels are converted to sufficiently large electrical DC voltages (4 V DC voltage from 2.25 mW optical power) using a stacked photodiode configuration. In the all-optical amplitude modulator, an 0.25 mW optical power in the control channel is sufficient to modulate the output power of the data channel by 15 dB. Also, we showed the results of all-optical controlled two-tap sequence detector for different modulation schemes OOK, PAM3, and PAM 4, at different settings. The tested symbol rate was 1.25 Gbaud; this can be easily increased by reducing the delay of the correlator tap. In the future, other all-optical processing circuits can be implemented using the same approach. We can accommodate more control signals for complex processing units by reducing the FSR of the wavelength demultiplexer and increase the number of channels. The loss of the charge depletion modulator can be reduced by shortening its length. This will increase the required driving voltage from the optical to voltage converter, achievable by cascading more photodiode in series.