Chip-scale optical interconnect using microLEDs

Abstract

In package intra-chip and/or inter-chip optical communications are provided using microLEDs and photodetectors mounted to integrated circuit (IC) chips and/or to transceiver dies associated with the IC chips. Light from the LEDs may pass through waveguides on or in a substrate to which the IC chips are mounted or which couple the IC chips.

Claims

1. An optical communications system for semiconductor chips on a common substrate, comprising: a plurality of first microLEDs on a first semiconductor chip, the first semiconductor chip electrically coupled to a substrate; a plurality of first photodetectors part of a second semiconductor chip, the second semi-conductor chip electrically coupled to the substrate; a plurality of waveguides, each of the plurality of waveguides coupling different ones of the plurality of first microLEDs with different ones of the plurality of first photodetectors; a plurality of second microLEDs on the second semiconductor chip; a plurality of second photodetectors part of the first semiconductor chip; and a further plurality of waveguides, each of the further plurality of waveguides coupling different ones of the plurality of second microLEDs with different ones of the plurality of second photodetectors; wherein the first semiconductor chip includes first drive circuitry for driving the plurality of first microLEDs and first receive circuitry for processing signals generated by the plurality of second photodetectors, and the second semiconductor chip includes second drive circuitry for driving the plurality of second microLEDs and second receive circuitry for processing signals generated by the plurality of first photodetectors; and further comprising a semiconductor package; wherein the plurality of first microLEDs and the first semiconductor chip are in the semiconductor package; and wherein the plurality of first photodetectors and the second semiconductor chip are in the semiconductor package.

2. The optical communication system of claim 1, wherein the first semiconductor chip comprises a first transceiver die and the second semiconductor chip comprises a second transceiver die, and further comprising: a third semiconductor chip electrically coupled to the first transceiver die; and a fourth semiconductor chip electrically coupled to the second transceiver die.

Description

BRIEF DESCRIPTION OF THE FIGURES

(1) FIG. 1 shows how a processor chip is currently packaged with high bandwidth memory (HBM) (Prior Art).

(2) FIG. 2 shows electronic ICs having microLEDs and detectors interfaced with an optical waveguide chip.

(3) FIG. 3 shows how three-dimensional waveguide chips can connect multiple ICs on different facets.

(4) FIG. 4 shows how the waveguide chip can also have metallization and vias to provide electrical connection to the silicon processor chip.

(5) FIGS. 5A-E illustrate integrated circuit chips coupled to a substrate, with LED based optical communications for communicating information between the integrated circuit chips.

(6) FIG. 6 shows how a microlens can increase the brightness of the beam and improve coupling to the optical waveguide.

DETAILED DESCRIPTION

(7) Embodiments in accordance with this invention use LEDs, microLEDs in various embodiments, for inter-chip communication, with in some embodiments the chips in communication being within a common package or on a common package substrate. Some embodiments use an array of microLEDs, such as blue LEDs on the order of 2 um×2 um each. In some embodiments a microLED is distinguished from a semiconductor laser (SL) as follows: (1) a microLED does not have an optical resonator structure; (2) the optical output from a microLED is almost completely spontaneous emission, whereas the output from a SL is dominantly stimulated emission; (3) the optical output from a microLED is temporally and spatially incoherent, whereas the output from a SL has significant temporal and spatial coherence; (4) a microLED is designed to be driven down to a zero minimum current, whereas a SL is designed to be driven down to a minimum threshold current, which is typically at least 1 mA. In some embodiments a microLED is distinguished from a standard LED by (1) having an emitting region of less than 25 um×25 um in some embodiments, of less than 10 um×10 um in some embodiments, of less than 5 um×5 um in some embodiments; (2) frequently having positive and negative contacts on top and bottom surfaces, whereas a standard LED typically has both positive and negative contacts on a single surface; (3) typically being used in large arrays for display and interconnect applications. In some embodiments the microLEDs are bonded to chips, in some embodiments wafer bonded to chips. For example, in some embodiments the microLEDs are bonded to a GPU and to memory stacks. The blue light, generated by the microLEDs, carries data which is received by detectors, for example either GaN detectors or silicon photodetectors/phototransistors. Memory modules, with similar bonded microLEDs connect to the GPU through 3D waveguides.

(8) Very high pixel per inch (PPI) microLEDs have been recently developed for various display applications, with sizes as small as 1 um reported in the literature. At this size, the LED has some properties that may be similar to a laser. The devices have high external quantum efficiencies matching those of lasers. Given that they are very small, they have small etendue—which means they can be coupled into small waveguides, very much like a laser. Though there is no stimulated emission, modulation speeds of a few GHz are possible, limited only by the RC time constant and carrier lifetime of the LED. Like carrier-induced waveguide modulators, faster response can be obtained by reversing the bias during the off-cycle. Faster response can also be obtained by driving the LED at high current density. Drive currents can be as low as 50 nA, with good external efficiency, and will realistically be limited by the bit error rate (BER) at the receiver at a given power.

(9) Bonding technology to attach ˜1 million microLEDs to silicon ICs has also been developed. Low capacitance can be obtained by an oxide-oxide bond with vias. A high temperature step expands the metals in the vias and fuses them together. Alignment between a source LED wafer and a target wafer (which is often a silicon wafer containing LED drive circuits) can be sub-micron using commercially available equipment.

(10) Blue GaN microLEDs typically have an emission peak at about 430 nm, a wavelength where silicon is highly absorbing and very fast photodetectors with small absorption lengths can be realized. Even lower noise operation can be obtained using an avalanche photodiode (APD) structure.

(11) Of course an LED does not have the coherence of a laser and emits light over a relatively wide optical spectrum. At long distances this can be a serious issue as any medium is dispersive and a broad spectral bandwidth will lead to a temporal spreading of the pulse with distance. However, this is not an issue as the application is typically about 1-5 Gb/s travelling a few centimeters. Numerically, the full width at half maximum (FWHM) of a blue microLED's optical spectrum centered at 430 nm might be about 20 nm, and the dispersion (dn/dλ) of fused silica about 0.08/um, so broadening would occur on a scale of many meters.

(12) The remaining problem is the optical interconnect between the chips. The simplest implementation would be to use a silica-on-silicon planar lightwave circuit (PLC) with mirrors or even angle polished edges, which is a mature technology. In some embodiments, SiO2, SiN, or SiON waveguide arrays may be used. In some embodiments polymer waveguides may be used. SiO2, SiN, or SiON waveguide arrays can be fabricated on a rigid planar substrate such as silicon or glass. Polymer waveguides can be fabricated on one of the rigid planar substrates previously mentioned, or on a flexible substrate such as mylar, supporting flexible waveguide arrays that are not constrained to lie in a plane. These waveguide arrays are typically fabricated using planar semiconductor techniques like photolithography, planar deposition, etching, and diffusion. In some embodiments, light is guided between chips by waveguides in a PLC, with 45 degree mirrors or angle polished edged used to couple vertically emitted light into horizontal waveguides. One could incorporate simple optical devices such as splitters into the PLC chip. Multiple layers of waveguides can be formed on top of each other, with mirrors and vertical waveguiding. An even more powerful implementation is to use 3D waveguides which can be created in glass using a number of techniques. A common method is by focusing a high power laser into bulk glass to generate defects that act as higher index material or microlenses. Arbitrary 3D waveguides can be formed in this way. These defects act as waveguides to carry the energy from the transmitter to the receiver. Alternative methods, such as 3D printing, can also be used to make dense optical waveguides. Even without waveguides, microlenses and mirrors can deflect thousands of beams simultaneously.

(13) These aspects are discussed through the use of the figures.

(14) FIG. 2 shows a simple implementation of an embodiment including aspects of the invention. In this case two chips are shown optically interconnected with optical waveguides. The leftmost chip is for example a CPU 211, and the rightmost chip is memory 213. The chips are electrically connected to a substrate or package from below using standard solder bumps or microbumps 215.

(15) Both chips contain microLEDs 217 and photodetectors 219. Though only a few microLEDs are shown, a typical application would have many tens of thousands of microLEDs, or even millions. The microLEDs are separately grown on a suitable “source” substrate 221 like sapphire, GaN, or silicon and then lifted off of the source substrate and bonded onto the “target” substrate or chip. Commercial equipment exists for this process, and in fact a photograph of a microLED on silicon is shown to the left. This technology was developed mostly for microLED displays where the silicon chip is a backplane to address the microLEDs. Typically solder is used for the bonding the microLEDs to the target substrate or chip, and there are often subsequent processing steps such as planarization and metallization to electrically connect the microLEDs to the CMOS silicon chip. The microLEDs would have a typical dimension of 2 microns×2 microns.

(16) The silicon chip would also have photodetectors fabricated on its surface. These can easily be made in silicon using a p-i-n or APD architecture. At wavelengths near 430 nm, silicon is highly absorbing, with an absorption coefficient of ˜5×10{circumflex over ( )}4/cm, and thus 1 micron of absorption length is sufficient to give very high quantum efficiency. Alternatively, detectors can be made in GaN or other materials and bonded to the silicon in the same way as the microLEDs.

(17) A waveguide structure 223 is aligned and attached to the assembly. The simplest implementation of this waveguide wafer is using well-established silica-on-silicon technology, with layers deposited by PECVD or FHD. These waveguides are generally formed on a silicon wafer by first growing a layer of thermal oxide 225, about 10 or 20 (or 2 to 5) microns thick, then depositing a higher index germanium-doped core 227, typically 4 (or 2) microns. Then this core is patterned laterally using photolithography, then a top cladding 229 is formed of silica. In the figure, the silicon substrate is not shown and the waveguide dimensions are greatly exaggerated. In addition to silica-on-silicon, other materials can be used, such as silicon oxynitride waveguides, polymer waveguides, or other dielectrics. The waveguide structures can be simple connections, or can contain splitters or possess other functionality. The waveguides can be single mode or multimode in the transverse directions. Given that the interconnect distance is very short, modal dispersion from a multimode waveguide will not limit the useful data rate of the optical signal in many cases.

(18) To couple light in and out of these waveguides various methods can be used. The simplest may be polishing the endface of the waveguide wafer at about 45 degrees. In this case light is reflected by the angled edge 231 into the waveguide. This is shown on the leftmost side of the figure. Alternatively, 45 degree mirrors 233 can be etched into the waveguide. These are shown schematically in the figure and a photograph of actual mirrors etched into silicon is shown to the right of the main figure. Note that there are multiple designs for this mirror. The core can be angled 45 degrees, or the light can exit the waveguide at a normal facet, then hit a second facet at 45 degrees. Light can also enter the waveguide using a grating. Instead of a mirror, a waveguide that is oriented normal to the bottom of the chip and then turns 90 degrees can also be used.

(19) The light is then coupled from the microLED into the waveguide, traverses across the chip(s) and is reflected downwards onto a photodetector. The connection can be across a single chip, or it can be across many chips.

(20) The small 2 um×2 um LEDs can be run at various current densities, depending on the optical loss between the LED and photodetector, the receiver sensitivity, and the required LED modulation speed. Current densities in the range of 5 A/cm{circumflex over ( )}2 to 10000 A/cm{circumflex over ( )}2 are typically useful, corresponding to drive currents of 200 nA to 400 μA.

(21) The waveguide structure shown in FIG. 2 is a 2D waveguide where the light is largely confined to the plane. However, waveguide structures can be realized that contain more than a single plane of waveguides. Waveguides can be formed on top of each other to make multiple planes. Perhaps more interestingly, there have been recent developments at fabricating 3D waveguide structures. An example is shown in FIG. 3, where multiple waveguides 311 shift their vertical orientation through a piece of glass 313. They can be fabricated with 3D printing, where layers are formed on top of each other. Alternatively, a high power beam is focused in the glass and pulsed, permanently altering the material's optical properties at the location where the beam waist is smallest. Multiple altered areas are connected, forming waveguides. These structures can then be used to connect to ICs at different facets of the glass. The top right image is a commercial product schematic of 3D waveguides formed in glass for connecting to a fiber array. The two lower figures show how these 3D waveguides can connect multiple ICs. For example, on the left image of FIG. 3, a GPU 315 is at one end of a block of glass 317, while memory 319a-d is connected to other facets, and 3D waveguides 321 connect the different chips. In some embodiments a GPU may include a up to 1 million optical inputs and outputs, with hundreds of memory chips optically connected to the GPU, each with tens of thousands of optical connections.

(22) By attaching an optical waveguide layer to the IC, we have potentially removed a surface that can be used for cooling. FIG. 4 illustrates an embodiment in which the optical waveguide chip 411 can also provide an electrical interface, leaving another surface 415 of the chip for cooling. In this way the optical waveguide chip becomes an interposer or a submount. The CPU/GPU chip 419 is now upside down with LEDs 421 and photodetectors 423 facing down. The optical waveguide layer has mirrors 425 that receive the light from the LEDs and direct them horizontally and back up again to a detector, as previously explained. However, this time, the chip is deep reactive ion etched (DRIE) to form through-chip vias 427 that can electrically connect to the CPU with standard solder bumps or microbumps. Even without vias, electrical signals 429 can be directed to the periphery of the chip and connected that way. It is simple to add metallization to the optical waveguide chip and add solder bumps or microbumps. The other side of the chip can now easily be cooled in a standard geometry. The waveguide can also connect to the other ICs.

(23) FIGS. 5A-E illustrate integrated circuit (IC) chips coupled to a substrate, with LED-based optical communications for communicating information between the integrated circuit chips. In most embodiments the LEDs are microLEDs. For convenience, only a single LED is shown in each of FIGS. 5A-E, in most embodiments a plurality of LEDs are present. In all of FIGS. 5A-E, a first IC chip 511a and a second IC chip 511b are both mounted to a substrate 513, on the same side of the substrate in FIGS. 5A-E. In some embodiments the substrate may be part of an integrated circuit package. In some embodiments the substrate may be an interposer. In some embodiments the substrate may be a bridge, with for example only part of each of the first IC chip and only part of the second IC chip mounted to the bridge. In some embodiments the first IC chip and the second IC chip are part of a multi-chip module. The IC chips are shown as having electrical connections via solder bumps or microbumps 515 to vias of the substrate. FIGS. 5A-E show the vias as through substrate vias (TSVs) 517, although in various embodiments the vias may lead to redistribution layers (RDLs) or other signal routing means in the substrates, and in some embodiments the substrate may not provide for electrical connections for the IC chips. An optical waveguide 519 is on the substrate, on the same side as the first IC chip and second IC chip. In some embodiments the optical waveguide is within the substrate, and in some embodiments the optical waveguide is on an opposing side of the substrate from the first IC chip and the second IC chip.

(24) In FIG. 5A, an LED 521 is on the first IC chip 511a. Light from the LED propagates through the optical waveguide 519 to a photodetector (not shown) on the second IC chip 511b. In some embodiments, and as illustrated in FIG. 5a, the LED has a height less than that of the solder bumps and/or microbumps, and is in a space between the first IC chip and substrate, with a height of the space determined by dimensions and/or properties of the solder bumps/microbumps. Driver circuitry for operating the LED, imposing data signals on light emitted by the LED, is part of the first IC chip. Receive circuitry for processing signals generated by the photodetector are part of the second IC chip. In many embodiments the photodetector is part of the receive circuitry. In some embodiments the photodetector is a discrete device mounted to the second IC chip, with the photodetector electrically coupled to the receive circuitry. In many embodiments, the optical waveguide also passes light from an LED (not shown) on the second IC chip, to a photodetector (not shown) on the first IC chip, with driver circuitry for the LED of the second chip part of the second IC chip and receive circuitry associated with the photodetector part of the first IC chip. As the combination of driver circuitry and receive circuitry may be commonly referred to as transceiver circuitry, for convenience FIGS. 5A-E simply show transceiver circuitry 523a,b. For FIG. 5A, the first IC chip includes the transceiver circuitry 523a and the second IC chip includes the transceiver circuitry 523b. The LED is shown as being on the part of the first IC chip having the transceiver circuitry 523a, although in some embodiments the LED may be differently positioned.

(25) For FIGS. 5B-E, the LED is not shown as being on the first IC chip. In FIG. 5B, the LED 521 is instead on the substrate 513, and the transceiver circuitry 523a (and 523b) is within the substrate. In FIG. 5B, the transceiver circuitry is shown as being within the substrate, and about a side of the substrate facing, and partially overlapping, the first IC chip. In various embodiments, however, the transceiver circuitry may be otherwise positioned within the substrate. The transceiver circuitry may be electrically coupled to the first IC chip by way of the electrical connections of the substrate and/or solder bumps/microbumps. The LED 521 is shown in FIG. 5B as being on the side of the substrate facing the first IC chip, at the location of the transceiver circuitry 523a. In some embodiments, however, the LED may be differently positioned. The transceiver circuitry 523b is correspondingly positioned and electrically coupled with respect to the substrate and the second IC chip.

(26) The embodiment of FIG. 5C differs from that of FIG. 5B in that the transceiver circuitry 523a,b is in a semiconductor die mounted in a cavity of the substrate. The LED may be mounted to the substrate, or, as illustrated in FIG. 5C, mounted to the transceiver semiconductor die.

(27) In the embodiment of FIG. 5D, the transceiver semiconductor die is mounted to the substrate, with the LED on the transceiver semiconductor die. In FIG. 5E, the transceiver semiconductor die is coupled to the substrate by one or more solder bumps or microbumps, with the LED on the transceiver die between the die and the substrate.

(28) Unlike a laser that emits light generally in the forward direction, the emission pattern from an LED is typically Lambertian. Addition of an optic, such as microlens, can tremendously improve the beam coupling to the waveguide. FIG. 6 shows an implementation of this approach, together with some simulation results of far-field angles with and without a microlens. In FIG. 6, a microLED 611 is on a CPU/GPU 613. A microlens 615 focuses at least some light emitted by the microLED onto a turning mirror 617. The turning mirror redirects the light by ninety degrees into an optical waveguide 619. The microlens may be part of a microlens assembly 621. In some embodiments the microlens assembly may include 1.5 um lenses at a 3.6 um pitch. With a FWHM of 23 degrees, as indicated by simulation, this is closely mode-matched to single mode fiber and can provide efficient coupling. In some embodiments, however, the lens is significantly larger than the LED, for example the lens may have a diameter at least two times a dimension of the LED in some embodiments, and at least eight times the dimension of the LED in other embodiments.

(29) Blue microLEDs may be considered to require a relatively large drive voltage compared to CMOS electronics, and suitable circuits may be formed in the VLSI to forward bias the LEDs. For faster response, a reverse bias voltage may be used to sweep the carriers out of the quantum wells in the LED. In some embodiments a higher voltage line is provided into and/or in the CMOS IC for the driver. Alternatively, GaN is an excellent electronic material, and a driver can be monolithically integrated with the LED. On the receive side, a detector made from Si or GaN detector can have very high bandwidth and can have very high quantum efficiency when used with a GaN LED emitting at a short wavelength. So various embodiments would have a very good receiver signal to noise ratio without requiring special receive circuitry.

(30) Although the invention has been discussed with respect to various embodiments, it should be recognized that the invention comprises the novel and non-obvious claims supported by this disclosure.