Faster data transmission rates are required to keep up with exploding data growth. While the present serialized single-lane rate is 100 Gb/s, past trends predict that this should reach 1 Tb/s rates in about a decade. How can we meet this technological challenge? Many see silicon photonics as part of the solution. On the other hand, as was the case for microprocessor chips, market forces can shift and we may momentarily give up on increasing the rate.

Here we argue that gap plasmons can provide the solution to this problem and the required technological ingredients already exist to vastly exceed Tb/s serialization in an integrated platform. Impressive progress has already been made on gap plasmon based high speed and high efficiency data processing, yet there has not been widescale adoption [1,2]. The vision of plasmonics playing a key role in these technologies is also not new, and the issues of loss were well recognized in those early works [3,4]. What is new are the additional technological elements that increase the capability of plasmonics.

1. Timescale of technology

The timescale for technologies to be adopted can be decades to centuries, depending on technology advancement and market forces. The first electric cars appeared in the 1800’s, yet only recently have they emerged as mainstream technology. In this case, many of the key technological elements already existed (or early versions of them did), yet the market was not ready. Equivalently in plasmonics, the Zenneck wave solution, which is equivalent to surface plasmons, was proposed in 1907 [5], yet the first commercial biosensors based on these waves became available in the 1990s, following initial demonstrations in the early 1980s [6]. With rapidly increasing demand for bandwidth, terabit Ethernet technology is highly desired by market leaders like Facebook and Google, the research has begun but the technology for it does not currently exist [7]. We are now at a point where the driving market forces may converge to connect the existing individual elements needed for ultrafast communications.

2. Technological elements

There have been recent demonstrations in plasmonics that provide a framework for future high speed data serialization. While many geometries are possible, we focus on the gap plasmon because it is most suited for electrical bias, as well as ease of fabrication and analysis. The following sections describe the plasmonic elements that could be combined for serialization/deserialization (SERDES) applications.

2.1 Electrodes

Probably the most obvious (and therefore easiest to overlook) advantage of gap plasmons is that they provide electrical contacts right up to the region of highest optical confinement. Doped silicon can provide good electrical conduction to a silicon-photonic platform, but the level of integration is still limited to half wavelength in the material. This is still quite large for 1550 nm, the low-loss fiber-optic wavelength, on the order of 250 nm. Gap plasmons can provide confinement down to the nanometer range (or even smaller), which is at the same scale as current electronic mass-fabrication. The electrical contact part is easy to overlook, but it plays the most important role in dense integration and connects the electronics without the usual optical-electrical-optical conversion detriments of high-cost, reduced speed, and low efficiency. Since the gaps are so small, the required voltage to achieve a high local field is also lower, so that efficient use of nonlinear processes is possible. Metals that have the highest free electron density and lowest scattering losses are preferred, giving the highest response for plasmonic resonances [8–10]. Aluminum is a particularly promising metal for wavelengths around 1.5 $\mu$m as it has a high $|\epsilon _m \mathfrak {R}\epsilon _m / \mathfrak {I} \epsilon _m|^2$ figure of merit with higher free electron density than gold and silver [8,9].

2.2 Photoconductive gaps as switchable gates

Plasmonic gap enhanced photoconductive antennas have recently emerged for terahertz source technologies [11–13]. These same photoconductive switches can be used to gate electro-optical interactions in ultrafast serialization technologies. They have extremely high bandwidth and remove heat (due to the metallization). They also have an extremely fast sweep-out time, which benefits by allowing for use of slow semiconductor materials (rather than more esoteric fast ones like low-temperature grown GaAs) [13]. It is possible that even silicon-based photoconductive switches will allow for terabit switching.

2.3 Schottky and photoemission detection

Plasmonic interfaces to semiconductors allow for sub-bandgap detection in the Schottky diode mode [14–18]. This allows detection of 1550 nm wavelength light while still using silicon-photonics. Photoemission detection is another approach to achieve below the bandgap detection [19]. The efficiency of these approaches is expected to improve with material processing, and they can be used to create carriers in photoconductive gaps as discussed in the previous subsection.

2.4 Nonlinear gap materials for encoding

Recent demonstrations have used second order nonlinear organic materials to encode data on a continuous wave (CW) laser [1]. Two dimensional materials also offer strong nonlinear response that may be incorporated in narrow gaps, and they are robust materials [20]. Because they are electrodes as well, the gaps can be used for loading the nonlinear material through dielectrophoresis [21] as well as poling the material if required [1]. Plasmonic organic hybrid modulators have been demonstrated with an in-device Pockel’s coefficient as high as 325 pm V$^{-1}$ near resonance [22].

2.5 Additional technologies

While tunneling based light emission and light-induced tunneling emission [23] are both ways of achieving light emission from plasmonic gaps, it is more likely that pulsed lasers will be an important way of providing a timing gate through photoconductive switching. Low cost femtosecond lasers, semiconductor lasers and fiber-based amplifiers are mature technologies to achieve the goal of switching and providing a source line for information to be encoded onto (particularly at the 1550 nm wavelength). Some challenges will exist in integrating plasmonics with silicon photonics, such as material compatibility (certain metals are not allowed) and ensuring that metal nanostructures can survive later processing steps [24,25]. In addition, there is a vast amount of nanofabrication capability and electronics capability that already exist, and these can be combined with gap plasmons to provide the required performance at the chip-level.

3. Impact of loss

When you have plasmonic gaps, typically the field, sweep out time, and nonlinear conversion scale inversely with the gap size (or a power of that); however, loss also will scale inversely with the gap size and can limit the functionality. While the loss generally scales inversely with length, the length is also reduced, and so overall, the best coupling occurs when the amount of loss equals the in-coupling efficiency [26]. The loss limit will often set in well before quantum tunneling or non-local effects. Indeed, experimentally nonlocal effects were not seen in gaps down to the sub-nanometer range in experiments [27–29].

Without loss, it is possible to make the gap as narrow as possible before tunneling sets in to get larger confinement, because the coupling is not fundamentally limited by the gap size. For example, nanoscale gaps of any size will transmit light at the single channel limit for the resonant case – it is gap size independent. With loss, the benefit of shrinking down should be traded off for the enhanced losses, so there is an optimal size for the gap. This usually occurs at the critical coupling point where scattering and absorption losses are matched [26]. For example, it has been shown that a slit of 14 nm was the optimal size for harmonic generation [30] and a 10 nm gap has been shown to be optimal for an application with plasmonic and excitonic resonances [31]. It is possible to go smaller to a few nanometers with optimization of the coupling design. This is an order of magnitude smaller than recent impressive demonstrations [1], so significant gains in integration and field enhancement are expected once this optimal size is achieved. Smaller gaps with lower voltages allow for reduction in losses in other parts of the circuitry, and lower heating since metals are good thermal conductors to remove heat – this has already played a beneficial role in THz photoconductive switching [32]. It is also larger than the size where expected tunneling or nonlocal effects (if they are present) will play a role.

4. Example configuration

In Fig. 1, an example configuration that employs the technology elements described above is shown. This is not a realistic embodiment that accounts for design in coupling, dispersion, manufacturability; but rather it shows how different elements that have already been demonstrated can be combined to produce new functionality. In particular, here we look at SERDES. At the top of the figure, an optical pulse from a femtosecond laser traverses photoconductive nanometric gaps that are biased by data pins operating at a lower clock rate (say 2.5 Gb/s). This makes use of THz response in photoconductive switching. The bias is constant for the time of the laser pulse, and this switches on the lines sequentially to encode data on a constant laser through a second order nonlinear process in the lower gap that uses a nonlinear material. The line is quickly switched off as the electrons are swept out. The pulsed laser operates at a 2.5 GHz repetition rate, and the time to travel between the photoconductive switching regions is equal to the output rate (1 Tb/s). This will require delay elements (not shown, however, Archimedean spirals are commonly used) because the distance travelled in 1 ps in a semiconductor is usually around 100 microns (too large for dense integration).

 

Fig. 1. Schematic of photoconductive switched serialization. This technique uses a pulsed laser to switch on parallel biases in series (top line), which modulate a nonlinear optical material in a 1550 nm data line (bottom). While only 8 parallel inputs are shown, it is envisioned that of the order of 400 inputs will be used in a Tb/s scheme. Inset, Modulator: Schematic of a plasmonic modulator using a gap filled with a nonlinear optical material. Inset, Photoconductive Switch: Image of a nanoplasmonic terahertz photoconductive switch on GaAs. Adapted with permission from [13]. Copyright (2012) American Chemical Society.

Download Full Size | PDF

In this way, a fan of parallel inputs can be serially combined at a higher rate on a single line. To achieve a Tb/s single-line encoding, 400 inputs each at 2.5 Gb/s can be used. Similar schemes can be used for deserialization that could make use of Schottky detection in gaps.

While loss has plagued plasmonics for many applications, there are high bit-rate, few-pass applications that can tolerate such loss. In particular, the SERDES application presented here is for serialization and so the main line need pass only a single plasmonic element to achieve serialization. While the loss on this line may be substantial, the ultimate out-coupled power will be in the milliwatt range.

5. Outlook

The imminent increases in data transmission provide an impetus to adopt gap plasmons as a core enabling technology. While many have envisioned this previously, the market conditions did not provide a strong enough motivation for transitioning away from established approaches. Gap plasmons have all the required technological elements already demonstrated – including the often overlooked feature of providing electrical contacts to the region of optical interaction. While early demonstrations used quite large gaps ($\simeq 100$ nm), much smaller gaps of a few nanometers will enhance the interaction, and allow for denser integration (due to the high effective index of the gap plasmon). Ultimately, loss does limit this approach, but the performance peaks for gaps of only a few nanometers. It is likely that combining forces with emerging silicon photonics will assist in bringing these technologies to market.