On-Chip Optical Nonreciprocity Using an Active Microcavity

Optically nonreciprocal devices provide critical functionalities such as light isolation and circulation in integrated photonic circuits for optical communications and information processing, but have been difficult to achieve. By exploring gain-saturation nonlinearity, we demonstrate on-chip optical nonreciprocity with excellent isolation performance within telecommunication wavelengths using only one toroid microcavity. Compatible with current complementary metal-oxide-semiconductor process, our compact and simple scheme works for a very wide range of input power levels from ~10 microwatts down to ~10 nanowatts, and exhibits remarkable properties of one-way light transport with sufficiently low insertion loss. These superior features make our device become a promising critical building block indispensable for future integrated nanophotonic networks.

. On-chip optical isolation and circulation using only one active whispering-gallery-mode (WGM) microtoroid resonator. The pump field is used to produce an effective gain ( F g in the forward propagation and B g in the backward) for the signal wave through optically pumping doped erbium-doped ions. 1 κ and 2 κ represent the coupling strengths of toroid-fiber 1 and toroid-fiber 2, respectively.
Light asymmetric transports are examined by measuring transmittance spectra in the forward (a) and backward (b) propagation configurations. in s stands for the amplitude of the input signal field.
is the real gain provided by optically pumping the doped erbium ions inside. These two sets of coupled equations (S1 and S2) provide the starting point for us to examine whether the signal transport lies in the reciprocal or nonreciprocal region.
Optical Isolation. In the case of optical isolation, we are interested in the quantity of isolation ratio (IR), which is defined as  (Figure 3b), or the input signal power ( Figure   S3). Although the one-cavity configuration makes the scheme elegant and simple, the gain-saturation nonlinearity actually brings rich physics into this system. Full theoretical evaluations with numerical simulations on the optical nonreciprocity of the device has been published elsewhere 2 . Further experimental results on optical isolation, especially on the insertion losses, will be continued below.

Optical Pseudo-Circulation (or Bidirectional Transmission).
In the present work, we choose ports 1, 2 and 3 ( Figure S1) to construct a three-port pseudo-circulator (or bidirectional transmission). The power flow direction is determined by coupling strengths. For simplicity, throughout the presentation (including the main text) we mainly concentrate on the following circulation scenario: if the signal enters into port 1, it drops out from port 2; but if the signal is launched from port 2, it does not come out of port 1, but instead exits from port 3. This situation can be easily fulfilled with asymmetric coupling ( 1 2 κ κ > ) in our current system. In terms of the scattering matrix, the signal outputs of this three-port pseudo-circulator is described by 2 without taking into account backscattering. Here to generalize, we mark signal launched through fiber 2 as forward propagation and signal launched through fiber 1 as backward propagation. Equation (S5b) indicates only one port works as a signal input port. Equation (S5) shows that if the signal is injected at port 3, the output will come out of port 2 instead of port 1. The lack of directional transmission 3 1 → implies the difference of our system from a conventional circulator. As a result, this drawback sets a limitation of the device in practical applications.
In the standard procedure, the performance of a circulator is characterized by computing the isolation ratios among different port combinations. To obtain a better quantitative characterization on the light circulation ability in our scheme, in this work we introduce the concept of directivity as another figure of merit to measure the signal-power flow in the direction of its strongest emission versus the opposite direction. By doing so, it leads to the definition of the forward directivity as 2 10 3 Maximum of Forward Directivity (dB) 10 log Maximum of to characterize the power flow direction as if the signal light is incident from port 1.
One can immediately deduce from Eq. S6 that in the ideal case without the existence of the backscattering, the forward directivity shall become infinity and the signal can only come out from port 2. (Recall that in this three-port pseudo-circulator, port 4 is not relevant in the analysis.) In reality, however, due to the surface roughness, the occurrence of Mie scattering will cause part of the reflected signal to be emitted from port 3 3 , which diminishes the forward directivity to be finite. In a similar way, we define the backward directivity Before moving to the next section, it is worth to mention that the concept of directivity has comprehensive applications in the fields of microwave circuits (such as distributed amplifiers) 4,5 and acoustics. We become also aware that in the characterization of the optical fiber networks, directivity has been commonly used to compute the ratio of the backscattered power at the other port to the input power in optical couplers 6 . It turns out that the application of the directivity in our work is very useful not only for evaluating quantitatively the isolation performance of the system, but also for guiding us to quickly tune the device into its optimal circulating behavior.

Further Information on Isolation Measurements
As described in the main text and the above section, high-contrast optical nonreciprocity in this system is obtained by asymmetric geometrical coupling when the system is operating in the gain-saturation regime. In light of the suggestions given in Ref. 7, all isolation measurements have been carefully implemented with the uses of optical switches (S1, S2 and S3 depicted in Figure 1d in the main text) to change the forward and backward configurations instead of manually realigning the optical paths. For more details, please refer to the schematic diagram of the experimental apparatus ( Figure 1d in the main text) and the above section. In the experiment, we have found that the ultimate isolation performance of the system is mainly limited by the power detection sensitivity of the adopted photodetectors.
To be more precise, as the fiber-cavity couplings become highly asymmetric, for a given input signal power its tiny backward transmission output basically falls below the commercial photodetector's noise level and is thus buried under the noise background. As a consequence, this results in a halted backward propagation and leads to near-perfect isolation performance. Such an extreme case has been experimentally approached and shown in Figure 2c in the main text. Since the optical power for the signal light is in the range of 10 nw ~10 µW and the optical quality factor of the cavity mode is around 10 6 , the thermal effect can be neglected in the silica microtoroid cavity 8 . This can further justified by the theoretical calculations as shown in Figures 2-4, in which the thermal effect is not taken into account.
In this section, we would like to continue further discussions on the isolation experiments by providing more information and results. In corresponding to Figure   3a-b given in the main text, here Figure S2 a and b supply the measured companying insertion losses in those two experiments. Note that the insertion loss is defined as 10 1 Insertion Loss (dB) 10 log Maximum of transmission ≡ × .

(S8)
Recall Figure 3a shown in the main text that by fixing 1 κ , as well as the input signal and pump laser powers, the isolation ratio increases along with enlarging the separation distance between the toroid and fiber 2. In such a case, as illustrated in Figure S2a, the insertion loss bears a similar trend. Specifically, the isolation ratio changes from near 0.5 dB up to 19.7 dB, and the insertion loss changes accordingly from 2.9 dB to 10.7 dB. For the situation of Figure 3b where 1 κ , 2 κ , and the injecting signal power are set to be constants, the isolation ratio increases from 3.0 dB up to 14.7 dB as the dropped pump power is increased. Interestingly, the insertion loss drops from 24.5 dB down to 7.4 dB in such a case (see Figure S2b). The reason for this insertion-loss reduction is obvious from the definition (S8) plus Eq. S1. That is, increasing the dropped pump power is equivalent to increase the original gain ( 0 g ), which allows the signal to receive gradually increased amplification. As a result, the insertion loss goes downward when coupling more dropped pump power into the active microcavity.
In addition to those two isolation experiments (depicted in Figure 2 a and b), we have also carried out another measurement for optical isolation as a function of the input signal power by setting all other parameters to be constants with the use of the second sample. The experimental data are plotted in Figure S3, which illustrates that the isolation ratio increases as coupling more signal intensity into the system. This change can be easily understood from the theoretical discussions presented above. That is, the difference of the effective gain between the forward and backward directions increases with the increase of the signal power. Using the theoretical model developed above and adding certain experimental details, we can quantitatively explain all the phenomena measured in Figures S2 and S3.
Let us make a short summary on the discussed on-chip optical isolation experiments.
From those implementations, it is now evident that the present microscale optical structure, compatible with the current COMS technology, is suitable for on-chip ultrasensitive, variable optical asymmetric transmission with the isolation ratio up to near infinity at the optical communication wavelength. The demonstrated optical isolation works in a broad operating scope for the input signal power ranging from ~10 nW up to ~10 µW. By manipulating the system parameters, the one-way light transport can be easily reversed for the potential application of an on-chip optical switch. More importantly, the exhibited sufficiently low insertion loss (for the isolation ratio up to 20 dB), a superior figure of merit, makes our simple scheme be a big step forward for the realization of a real, practical on-chip optical directional transmission device for integrated photonic networks. In addition, limited mainly by the power detection sensitivity of the adopted photodetectors and the insertion loss in forward transmission measurement, in the present experiment the recorded lowest signal power of ~10 nW with an appreciable isolation ratio of about 20 dB would be very challenging without the use of a resonant structure.

Further Information on Pseudo-Circulation Measurements
In the proposed chip-based optical circulators with ring resonators and photonic crystals 9,10 , both approaches are still based upon the magneto-optically induced frequency splitting between the clockwise and the counter-clockwise traveling modes.
Different from those proposals, in our current work we exploit the gain-saturation nonlinearity in an active microtoroid resonator under asymmetrical coupling strengths between the microtoroid and two tapered optical fibers to achieve bidirectional transmission. Although a real optical circulator is also one type of one-way light propagation devices, it requires more constraints than a simple optical isolator. In particular, the directional circulation performance of our scheme can be further characterized via measuring the defined directivity. As a typical example, Figure 4b in the main text illustrates the trends of the forward and backward directivities as a function of the separation distance between the microcavity and fiber taper 2. It shows that as the separation distance increases, the forward directivity drops from 9.5 dB down to 2.7 dB while the backward directivity grows from 6.7 dB up to 16.8 dB. In corresponding with Figure 4b in the main text, Figure S4 further provides detailed information on the variations of the companying forward and backward insertion losses. It is apparent from Figure S4 that the insertion losses for the forward and backward transmissions abide by opposite behaviors. Specially, as the separation distance between the microcavity and fiber taper 2 increases, the forward insertion loss grows from 1.3 dB to 9.1 dB while the backward insertion loss deceases from -1.8 dB to -3.1 dB. Such changes can be understood as follows: for the backward propagation configuration, the signal light is launched from port 2. Increasing the distance between the microcavity and fiber taper 2 effectively reduces 2 κ , and thus leads to the growth of the circulating optical power inside the microcavity. This results in more transmitted optical power from port 2 to port 3 and, consequently, leads to the reduction of the insertion loss in the backward configuration. On the other hand, for the forward direction, the signal light is launched from port 1. Therefore, reducing 2 κ results in a decrease of the circulating optical power inside the microcavity, and thus leads to the reduction of the dropped signal power from port 2.
As a result, the insertion loss increases in the forward configuration. The negative sign appearing in front of the backward insertion loss is a direct signature of the amplification that the signal light experiences in the path traversing from port 2 to port 3 in the backward transmission configuration.
In Figure S5, the optical bidirectional-transmission performance is also experimentally studied as a function of the dropped pump power by fixing all the other parameters. The experimental data indicate that along with increasing the dropped pump power, both forward and backward directivities get improved before merging into certain steady (saturated) value(s). The essential physics behind this measurement relies on the fact that as the pump power is gradually increased, 0 g first increases and then becomes saturated into a constant.
In the end, we want to make a few remarks on the appearance of a doublet structure shown in the subplot 2 3 → in Figure 4a of the main text. First of all, one should be cautious that this doublet has a completely different physical origin from that due to the scattering-induced mode splitting as seen in Figure 2a