Method and apparatus of surface-incident, plasmon-enhanced multiple quantum well modulators and optical coupling thereon

Abstract

An optical interconnect system has first and second waveguides each with wedge-shaped cross-section at a first end, disposed over an optical modulator. The optical modulator is a surface-plasmon multi quantum well (SP-MQW) modulator, the first waveguide an input waveguide and the second waveguide configured an output waveguide. In embodiments the SP-MQW modulator has multiple semiconductor layers disposed atop a lower metal layer between 10 and 300 nanometers thick and configured such that incident light is reflected at the lower metal layer unless a voltage is applied to the semiconductor layers, when incident light is coupled into a surface plasmon mode in the lower metal layer.

Claims

1. An optical interconnect system comprising a first and a second waveguide each having wedge-shaped cross-section at a first end, the first and second waveguide attached to a surface of an integrated circuit with the first end disposed over an optical modulator; wherein the optical modulator is a surface-plasmon multi quantum well (SP-MQW) modulator, the first waveguide configured as an input waveguide to the SP-MQW modulator and the second waveguide configured as an output waveguide from the SP-MQW modulator; and wherein the SP-MQW modulator comprises a plurality of semiconductor layers disposed atop a lower metal layer between 10 and 300 nanometers thick and configured such that incident light is reflected at the lower metal layer unless a voltage is applied to the semiconductor layers, whereupon incident light is coupled into a surface plasmon mode in the lower metal layer.

2. The optical interconnect system of claim 1 wherein the lower metal layer comprises at least one of the group consisting of copper (Cu), Titanium (Ti), Gold (Au), Aluminum (Al), Nickel (Ni), and Cobalt (Co).

3. The optical interconnect system of claim 2 wherein the lower metal layer lies atop a lower dielectric layer.

4. The optical interconnect system of claim 3 further comprising an upper dielectric layer atop the semiconductor layers.

5. The optical interconnect system of claim 3 further comprising an upper dielectric layer atop the upper metal layer, the upper and lower dielectric layers having different refractive index.

6. The optical interconnect system of claim 1 further comprising an upper metal layer deposited atop the semiconductor layers, the upper metal layer between 10 and 100 nanometers thick.

7. A surface-incident, plasmon-enhanced, multiple quantum well optical modulator comprising: a multiple quantum well structure comprising at least two layers of two different semiconductor materials; at least one metallic layer adjacent to said multiple quantum well structure; at least one dielectric layer on top of said multiple quantum well structure; at least a second dielectric layer underneath said multiple quantum well; said second dielectric layer having a different refractive index from said first dielectric layer; said metal layer being also adjacent to one of said dielectric layers; where light is incident through one of said dielectric layers with a higher refractive index on one surface of said multiple quantum well; the incidence angle being large enough to allow total internal reflection at a second surface of said multiple quantum well when no voltage is applied on said modulator structure; while applying a voltage on said modulator structure changes the refractive index and absorption coefficient of said multiple quantum wells and couples the incident light into at least one mode selected from the group consisting of a surface plasmon mode propagating on the surface of said metal layer and coupled modes between said surface plasmon mode and metal/MQW/metal guided modes.

8. A method of forming a surface-incident, plasma-enhanced multiple quantum well optical modulator structure comprising: forming a multiple quantum well structure comprising at least two layers of two different semiconductor materials; forming at least one metallic layer adjacent to said multiple quantum well; forming at least one dielectric layer on top of said multiple quantum well; forming at least a second dielectric layer underneath said multiple quantum well; said second dielectric layer has a different refractive index from said first dielectric layer; said metallic layer being also adjacent to one of said dielectric layers; where light is incident through one of said dielectric layers with a higher refractive index on one surface of said multiple quantum well; the incidence angle being large enough to allow total internal reflection at a second surface of said multiple quantum well when no voltage is applied on said modulator structure; while with a voltage applied on said modulator structure refractive index changes and couples a part of the incident light into at least one mode selected from a surface plasmon mode propagating on the surface of said metal layer and coupled modes between said surface plasmon mode and metal/MQW/metal guided modes, thereby reducing the internal reflection.

Description

BRIEF DESCRIPTION OF THE FIGURES

(1) FIG. 1 illustrates an on-chip and intra-chip optical interconnection system.

(2) FIG. 2 illustrates PRIOR ART transmission contrast ratio vs. wavelength at different driving voltages for state-of-the-art, conventional surface-incident electro-absorption modulators (EAMs).

(3) FIG. 3 illustrates changes in absorption coefficient and refractive index n in Al0.32Ga0.64As (10 nm)/GaAs (10 nm) quantum wells (QWs) under 70 kV/cm electric field.

(4) FIG. 4 illustrates schematic Surface Plasmon Enhanced (SP)-Multiple Quantum Well modulator (MQW) (SP-MQW) structures (right) and simulated reflectance as a function of incidence angle under 70 kV/cm field (left) using the Al0.32Ga0.64As (10 nm)/GaAs (10 nm) MQW structure;

(5) FIG. 5 is a cross sectional diagram of an embodiment of the Multiple Quantum Well MQW/metal SP-MQW structure.

(6) FIG. 6 illustrates dependence of IL and ER of the SP-MQW modulator on incident beam divergence angles.

(7) FIG. 7 is a cross sectional diagram of an alternative P-I-N MQW embodiment.

(8) FIG. 8 is a cross sectional diagram of an alternative Schottky-barrier embodiment.

(9) FIG. 9 is a cross sectional diagram of an alternative field-effect embodiment.

(10) FIGS. 10-11 further detail an embodiment of the tapered facet coupling of a single optical mode to the SP-MQW. The schematic cross-section structure is shown in FIG. 10, while an optical simulation is shown in FIG. 11. A coupling efficiency greater than 85% can be achieved.

(11) FIGS. 12 and 13 shows how extinction ratio and insertion loss vary with angle of incidence.

DETAILED DESCRIPTION OF THE EMBODIMENTS

(12) We disclose a Surface-incident, Plasmon-enhanced Multiple Quantum Well modulator (SP-MQW) and its optical coupling scheme. These modulators will be used in computers and data centers to achieve highly energy-efficient optical interconnects to transport information from one computer chip to another.

(13) Conventional surface-incident EAM modulators suffer from limited extinction ratio and/or high driving voltage due to the limited absorption per unit thickness. For example, even though a large absorption coefficient change of absorption coefficient >5,000 cm.sup.1 can be obtained in Al.sub.0.32Ga.sub.0.64As (10 nm)/GaAs (10 nm) QWs under a 70 kV/cm electric field, to achieve 7 dB extinction ratio the thickness of the MQW stack has to be greater than 1.5 m. This leads to a driving voltage of 70 kV/cm1.5 m10 V, too high for complimentary metal-oxide semiconductor (CMOS) circuitry of modern high-performance processor integrated circuits to handle without relaxed design rules and, in some cases, extra processing steps for high voltage transistors. Modern CMOS circuits require driving voltages are below 1.5 V. The high driving voltage also leads to high power consumption, which is proportional to voltage squared. Furthermore, even with such a high driving voltage, the contrast ratio is still very limited in state-of-the-art surface-incident EAMs, as illustrated in FIG. 2 where wavelength dependence of contrast 202 at 2V is compared with wavelength dependence at higher voltages, including at 8V 204. In the prior surface-incident EAM, of FIG. 2, the extinction ratio did not exceed 2 dB even at 8V driving voltage in this case.

(14) To address the challenges in low driving voltage and high extinction ratio, in this invention we synergistically utilize both refractive index n and absorption coefficient (FIG. 3) in a MQW modulator via surface plasmon coupling (FIG. 4). As shown in FIG. 3, according to Kramer-Kronig relation, the change in absorption coefficient upon the application of an electric field/voltage will also induce a significant change in refractive index n. In this embodiment, we use Al.sub.0.32Ga.sub.0.64As (10 nm)/GaAs (10 nm) QWs as an example. The under an applied electric field is induced by the quantum confined Stark effect (QCSE). While conventional EOMs typically achieves a n of 0.001 under an applied electrical field of 100 kV/cm, here we can achieve a much larger n=0.08 at =850 nm (FIG. 3) under the 70 kV/cm field due to the large in the wavelength range nearby, almost 100 greater than conventional EOMs.

(15) The corresponding device structure 300 of the SP-MQW is shown in FIG. 5 The MQW has two different types of semiconductor materials with different band gaps.

(16) In an embodiment, incident light from a laser diode, in a particular embodiment a VCSEL laser, that may be located on or off a digital integrated circuit (chip) is coupled into an incident-light waveguide 302 (FIG. 5). Incident light waveguide 302 routes the incident light, symbolized by arrows 304, to a wedge coupler 303 diverting it downwards through transparent a transparent dielectric 306 into the SP-MQW device 308 at an angle of incidence .sub.1. The MQW device has alternating layers of first semiconductor 312 and second semiconductor 314.

(17) Suppose semiconductor 312 is the quantum well (QW) while semiconductor 314 is the barrier to confine the electrons in semiconductor 312, then the band gap of semiconductor 312 has to be smaller than that of semiconductor 314. The thickness of each layer in the MQW structure is less than 50 nanometers (nm). In one embodiment, semiconductor 312 is Gallium Arsenide (GaAs) and semiconductor 314 is an Aluminum Gallium Arsenide (Al.sub.0.32Ga.sub.0.64As). In another embodiment, semiconductor 312 is Indium Gallium Arsenide (InGaAs), Indium Gallium Arsenide Phosphide (InGaAsP), Aluminum Gallium Arsenide (AlGaAs) or other alloys having similar properties; semiconductor 314 can be Aluminum Gallium Arsenide (AlGaAs) of other dopings and/or compositions. In an embodiment, the alternating semiconductor layers 312, 314 of the MQW total thickness 316 about 100-900 nm in thickness and in a particular embodiment 200 nm. In other embodiments, the MQW structure is substituted with a uniform semiconductor thin film instead of multiple layers, and the under an applied electric field is induced by Franz-Keldysh effect.

(18) A thin metal layer 318 of less than 150 nm thickness 319 is formed at the bottom of the MQW. In one embodiment, the metal is copper (Cu), a metal that can undergo plasmon interactions with photons. In other embodiments, the metal is Titanium (Ti), Gold (Au), Aluminum (Al), Nickel (Ni), Cobalt (Co), or another metal alloy that can undergo plasmon resonance interactions with photons. Dielectric layer 306 has refractive index n.sub.1 and is formed on top of the MQW, while another dielectric layer 320 with refractive index n.sub.2 is formed under the metal layer. Here n.sub.1>n.sub.2. The dielectric materials are chosen from but not restricted to polymers, Silicon Dioxide glass (SiO2), Silicon OxyNitride (SiOxNy), Silicon Nitride (SiNx), and air.

(19) During device operation, light is incident on the top surface of the MQW at an incidence angle of .sub.1>arcsine(n.sub.2/n.sub.1). This angle 1 is also within 60 degrees of the incidence angle for optical coupling into the surface plasmon (SP) mode of metal layer 318, Op. That is, |1p|<60 degrees. Note that Op is determined by the refractive indices and thicknesses of the MQW, the metal layer, dielectric layer 306, and dielectric layer 320. When the refractive index of the MQW is changed through operation of n under an applied electric field, Op can be shifted to .sub.1 to achieve optical extinction by coupling incident light into the SP mode 322 that propagates along the metal surface rather than being reflected. In one embodiment in FIG. 4, .sub.1=77.4 degrees; p.sub.1=72.5 degrees without the applied electric field, and it is shifted to p.sub.2=.sub.1=77.4 degrees when an electric field of 70 kV/cm is applied on the MQW structure. Incident light 304 is largely reflected as output light 324 by the semiconductor 314-metal 318 interface without the electric field, but with the electric field generation of surface plasmons from photons absorbs energy and reduces reflected output light 324. Generation of these plasmons increases extinction of output reflected light 324, thereby increasing the extinction ratio ER.

(20) Output reflected light 324 is captured by a second wedge coupler 326 and diverted into output waveguide 328, from whence it may be routed through optical interconnect, either chip-chip or chip from module to module to chip, to an optical receiver, not shown.

(21) The SP-MQW device is built atop a digital integrated circuit. Beneath the lower dielectric layer 320 are interconnect and metallization layers 341, and semiconductor layers 343, of the underlying high performance digital CMOS integrated circuit.

(22) There are two major benefits of the disclosed SP-MQW modulator structure:

(23) (1) Without the applied field, the incident light essentially experiences total internal reflection at the MQW/metal interface, with little or no coupling to the lossy surface plasma mode, leading to less than 1.5 dB insertion loss (IL) (352 in FIG. 4, where the reflectance is 80%).

(24) (2) With the applied electric field inducing a large refractive index change in the MQW (n0.08); the light is largely coupled to a surface plasma mode (SP) mode that propagates along the surface of the metal. Correspondingly, the reflectance is minimized and a high ER is achieved 354 in FIG. 4, where the reflectance is 0. This way, we use the plasmonic loss to our advantage. Most importantly, the optical decay tail in the MQW is <40 nm, meaning that even a 100 nm-thick MQW stack can fully confine the SP mode and induce a large ER upon SP coupling, thereby allowing a low driving voltage of 0.7 V to achieve 70 kV/cm electric field (=0.7 V/100 nm). This <1 V driving voltage is fully compatible with low-power CMOS driving circuitry such as that on modern, high performance, processor chips. Compared to the existing art, the driving voltage is decreased by 10, the energy consumption by 100, and the extinction ratio increased from 2 dB to >15 dB (FIG. 6) where unbiased reflection 360 shows a small injection loss, and SP-enhanced reflection 362 shows a much higher 362 injection loss for single or multi-mode waveguide operation.

(25) In the preliminary design modeling shown in FIG. 4, the reflectance is reduced from 80% to 0 at the optimal incidence angle under the applied field, with an angular divergence of .sub.11. This angular divergence is easily satisfied by single-mode incident light which has .sub.0<1 divergence. As shown in FIG. 6, a very high extinction ratio exceeding 16 dB can be achieved for a beam divergence of .sub.0=10 at <1.5 dB insertion loss. The high extinction ratio suggests that the modulator design is also compatible with advanced modulation formats such as 4-level Pulse Amplitude Modulation (PAM-4) in addition to simple on-off keying. As the divergence angle increases, the insertion loss remains almost the same while the extinction ratio decreases. Considering that well-designed multi-mode waveguides can achieve 1<5, an ER >6 dB is still readily attainable, much larger than the state-of-the-art shown in FIG. 1 (ER=2 dB). Considering a 55 m.sup.2 device area and the 0.7 V driving voltage, the energy efficiency is as high as 3.6 fJ/bit, more than 100 better than existing art. This can be further increased with optimized design of the MQW/metal/dielectric stack. The RC limited 3 dB bandwidth is as high as 100 GHz assuming 50 load, which can be increased further since the load resistance can be less than 50 for optical interconnects.

(26) In a particular embodiment, in order to further enhance ER, an upper copper layer 330 is added between upper dielectric 306 and semiconductor layers 312, 314 of the MQW device. In this embodiment, the upper metal layer 330 is between 10 and 100 nanometers thickness 331, and in a particular embodiment 30 nanometers. Similarly, the lower metal layer 318 is between 10 and 200 nm, and in a particular embodiment 50 nm, thick. In an embodiment, both metal layers 318 and 330 are copper, in an alternative embodiment lower metal layer 318 is Cu.sub.3Ge, while Au is a good ohmic contact metal for MQWs.

(27) In an alternative embodiment 400 the top 402 and bottom 406 of the MQW structure is different diffused (FIG. 7); one is n-type and the other is p-type, while the center of the MQW is intrinsic 406, forming a P-I-N structure. Remaining features of alternative embodiment 400 are described with reference to FIG. 5. An electric field is established by applying a reverse bias across the P-I-N junction MQW structure.

(28) In another embodiment 450 (FIG. 8) with doped semiconductor layers 452, 454, the doped semiconductor electrode at the bottom is omitted, while the Schottky barrier between the metal 456 and the MQW is used to establish an electric field under reverse bias. Remaining features of alternative embodiment 450 are described with reference to FIG. 5.

(29) In still another embodiment 500 (FIG. 9), a thin dielectric insulator layer 506, such as SiO.sub.2, Al.sub.2O.sub.3, or another oxide layer, with a refractive index of n.sub.3 is inserted between the metal and the MQW N and P regions 502, 504 to form a metal-oxide-semiconductor (MOS) structure. An electric field is established in the MQWs under depletion mode the MOS structure (reverse bias). In this case, the dielectric layer n.sub.2 may be omitted if n.sub.3<n.sub.1, as shown

(30) An embodiment of optical coupling to the SP-MQW and integration with the photonic datalink system is schematically shown in FIG. 1. The SP-MQW couples to waveguides of the device with high efficiency, as illustrated in FIG. 10-11. Here CMOS electronic integrated circuits (ICs) are bonded on chip carriers, which are mounted on printed circuit boards. Each CMOS electronic chip is connected to an optical chip/bridge on its edge. On the photonic chip/bridge, light emitted by a vertical cavity surface emission laser (VCSEL) is connected to a waveguide via a 45 degree facet (see the box on the right). In one embodiment, the waveguide is made of polymer material. In other embodiments, the waveguide is SiON, SiNx, Si or other dielectric materials. The light propagates in the waveguide, and when approaching the SP-MQW it is coupled to the modulator at a desirable incidence angle (.sub.1 in FIG. 8) via tapered facet couplers as shown in FIGS. 5 and 10-11. When no electric field is applied on the SP-MQW, the light incident on the SP-MQW is reflected via total internal reflection through a second tapered facet into the output waveguide. When an electric field is applied, the incident light is coupled to the surface plasmon mode on the surface of the metal layer due to the change in refractive index and the absorption coefficient in the MQW structure. Consequently, there is no reflected light coupled to the output waveguide. Therefore, the modulation of light intensity and optical encoding is implemented. Note that the SP-MQW structure can also be used as a photodetector (PD) with a similar coupling scheme. Both the SP-MQW and the PD are electrically connected to the silicon underlying CMOS chip.

(31) At the edge of the photonic chip/bridge, the waveguide on the photonic bridge is coupled to a flexible waveguide ribbon via 45 degree reflector facets and microlenses (see the box on the top left of FIG. 3) so that optical signals can be transmitted from one chip to another.

(32) In a particular Schottky embodiment with the optional upper metal, as shown in FIG. 12, optimum angle of incidence is 41 degrees.

(33) At incident angles smaller than the SP resonances, a low IL of 1.3-1.6 dB and a large ER of 5-7.5 dB can be achieved. Note that the 50 kV/cm difference in the high (70 kV/cm)/low field (20 KV/cm) corresponds to a low voltage swing of V.sub.pp=1 V, thanks to the small thickness of the MQW layer (200 nm). Therefore, low IL and high ER can be achieved at a low driving voltage. FIG. 3b shows that the device could also work in some regimes with a larger incidence angle than the SP resonance. Therefore, this design also offers some flexibility and tolerance to angular divergence of the incident beam.

(34) Extinction ratio and insertion loss of the double metal layer 30 nm Cu/200 nm MQW/50 nm Cu structure are discussed with reference to FIG. 12 at two different incident angle ranges. (a) 39-40.6 degrees and (b) 42-56 degrees are illustrated in FIG. 13.

(35) In an alternative embodiment, operation in the range illustrated in either range (a) or range (b) of FIG. 13 is used.

(36) Key Advantages Compared to Existing Arts

(37) Compared to conventional structures without surface plasmon assistance, key novel features are summarized in the table below. The performance is drastically better than prior surface incident modulators.

(38) TABLE-US-00001 Key Performance SP-MQW modulator Prior Surface-incident Parameters in this Invention Modulators Extinction Ratio >15 dB <3 dB (the higher the better) Driving Voltage 0.7-1 V >5 V (the lower the better) Energy Consumption 1-10 fJ/bit 500-1000 fJ/bit (the lower the better)

(39) We envision that the proposed technology has great potential to be deployed in the form of electronic-photonic co-packaged chipsets for high bandwidth-demanding chip-to-chip as well as chip-to-connector communications. High performance server systems and hyper-scale data center switching systems are among first potential adopters of our technology. The trend of transitioning from system faceplate pluggable optical modules to on-board optical modules is steadily progressing from IBM's pioneering attempt in their P775 supercomputer systems to the fast-moving and influential COBO (Consortium for On-Board Optics) founded and supported by a number of major players in the industry. With the ever-growing demands for bandwidth density within a manageable power budget, the next logical step is clearly to move optics even closer to electronics by realizing electronic-photonic co-packaging in one chipset.

(40) We are convinced that the key components adopted in our interconnect scheme, such as the disclosed SP-MQW modulator, provides solid cost and manufacturability advantages of our proposed technology in a commercial setting. Based on prior cost analysis of hybrid integrated optical transceiver module, the two major cost factors are active optoelectronic component and sub-assembly costs as well as system packaging and assembly, contributing to 54% and 36% respectively of the final module cost. Our proposed technology can significantly reduce costs associated with both components and packaging. The use of surface normal optical chips such as VCSELs, SP-MQW modulators and PDs are made by semiconductor processes on 6 GaAs wafers. In the fiber optics industry, by taking advantage of a nearly $10B GaAs IC ecosystem (the 3rd most maturely developed IC ecosystem after Si and SiGe), GaAs-based chips and solutions (mainly for short-reach applications) claim nearly one order of magnitude lower cost than their InP counterparts and already achieved $1/Gb/s half a decade ago. The use of surface normal devices is the key to reaching the $0.1/Gb/s target. Costs of surface normal chips (e.g. VCSELs) have consistently been a fraction of those of their waveguide coupled counterparts (e.g. DFB) over the past 20 years due to larger wafer size (6 GaAs vs. 2-3 InP), much higher processing yield, and significant cost savings on testing (wafer level vs. individual edge tests). It is also cheaper than Si photonics which still requires hybrid integration of expensive DFB lasers. High-volume pricing of commercial VCSELs is now below $0.05/Gb/s (even lower with one DC VCSEL powering multiple channels) and the costs of MQW modulators and PDs are minimal due to simpler epi-structures. The polymer waveguides can be made via an industrially validated low-cost roll-to-roll imprint process. The material platforms, fabrication methods and packaging scheme define a clear path to the $0.1/Gb/s cost target.

(41) Changes may be made in the above methods and systems without departing from the scope hereof. It should thus be noted that the matter contained in the above description or shown in the accompanying drawings should be interpreted as illustrative and not in a limiting sense. The following claims are intended to cover all generic and specific features described herein, as well as all statements of the scope of the present method and system, which, as a matter of language, might be said to fall therebetween. It is also anticipated that steps of methods may be performed in an order different from that illustrated and still be within the meaning of the claims.