Major silicon photonics breakthrough could allow for continued exponential growth in microprocessors

October 8, 2013

A microchip that contains both photonics and electronics is tested at CU-Boulder researcher Milos Popovic’s lab (credit: Casey Cass/CU-Boulder)

Researchers at the University of Colorado Boulder, the Massachusetts Institute of Technology, and Micron Technology Inc. have developed a new technique in silicon photonics that could allow for exponential improvement in microprocessors to continue well into the future.

The technique allows microprocessors to use light instead of electrical wires to communicate with transistors on a single chip, a system that could also lead to energy-efficient computing.

CU-Boulder researcher Milos Popovic, an assistant professor of electrical, computer and energy engineering, and his colleagues created two different optical modulators — structures that detect electrical signals and translate them into optical waves — that can be fabricated within the same processes already used in industry to create today’s state-of-the-art electronic microprocessors.

Microelectronics also are limited by the fact that placing electrical wires that carry data too closely together can result in “cross talk” between the wires.

In the last half-dozen years, microprocessor manufacturers, such as Intel, have been able to continue increasing computing speed by packing more than one microprocessor into a single chip to create multiple “cores.” But that technique is limited by the amount of communication that then becomes necessary between the microprocessors, which also requires hefty electricity consumption.

Advantages of using photons instead of electrons

A silicon wafer containing the photonic-electronic microchips designed by the research team, which includes scientists from CU-Boulder, MIT, Micron and UC Berkeley (credit: Milos Popovic/CU-Boulder)

Using light waves instead of electrical wires for microprocessor communication functions could eliminate the limitations now faced by conventional microprocessors, Popovic said.

Optical communication circuits, known as photonics, have two main advantages over communication that relies on conventional wires: using light has the potential to be brutally energy efficient, and a single fiber-optic strand can carry a thousand different wavelengths of light at the same time, allowing for multiple communications to be carried simultaneously in a small space and eliminating cross talk.

Optical communication is already the foundation of the Internet and the majority of phone lines. But to make optical communication an economically viable option for microprocessors, the photonics technology has to be fabricated in the same foundries that are being used to create the microprocessors.

Photonics have to be integrated side-by-side with the electronics in order to get buy-in from the microprocessor industry, Popovic said.

“In order to convince the semiconductor industry to incorporate photonics into microelectronics you need to make it so that the billions of dollars of existing infrastructure does not need to be wiped out and redone,” Popovic said.

Integration with existing infrastructure

Last year, Popovic collaborated with scientists at MIT to show, for the first time, that such integration is possible. “We are building photonics inside the exact same process that they build microelectronics in,” Popovic said. “We use this fabrication process and instead of making just electrical circuits, we make photonics next to the electrical circuits so they can talk to each other.”

A digital rendering of an optical modulator that can be fabricated using the same process that industry is already using to make today’s state-of-the-art electronic microprocessors (credit: Milos Popovic/CU-Boulder)

In two papers published in Optics Letters with CU-Boulder postdoctoral researcher Jeffrey Shainline as lead author, the research team refined their original photonic-electronic chip further, detailing how the crucial optical modulator, which encodes data on streams of light, could be improved to become more energy efficient.

That optical modulator is compatible with a manufacturing process — known as Silicon-on-Insulator Complementary Metal-Oxide-Semiconductor, or SOI CMOS — used to create state-of-the-art multicore microprocessors such as the IBM Power7 and Cell, which is used in the Sony PlayStation 3.

The researchers also detailed a second type of optical modulator that could be used in a different chip-manufacturing process, called bulk CMOS, which is used to make memory chips and the majority of the world’s high-end microprocessors.

Vladimir Stojanovic, who leads one of the MIT teams collaborating on the project and who is the lead principal investigator for the overall research program, said the group’s work on optical modulators is a significant step forward.

“On top of the energy-efficiency and bandwidth-density advantages of silicon-photonics over electrical wires, photonics integrated into CMOS processes with no process changes provides enormous cost-benefits and advantage over traditional photonic systems,” Stojanovic said.

The CU-led effort is a part of a larger project on building a complete photonic processor-memory system, which includes research teams from MIT, Micron Technology, and the University of California, Berkeley.

So when will we see products on the market?

“This innovation could enable optical processor to memory interconnects in supercomputers within 5 years, and likely within consumer electronics like game consoles, cell phones, as well,” Popovic told KurzweilAI. “Chips enabled with optical processing could also impact an array of other areas, including medical imaging, advanced analog signal processing like analog to digital conversion, etc. The bottom line is that optics is becoming an essential part of advanced microelectronics as performance continues to scale.

“The uniqueness of this innovation in comparison to all other work in the field is that it allows the construction of active photonic (light) circuits in unmodified advanced CMOS processes already available in billion dollar foundries. While there are now other research groups working on monolithic integration, such as at IBM, no other approach uses an entirely modified CMOS process as ours does.

“Our approach allows designers to build photonics inside advanced CMOS electronics today, using the standard state of the art electronics infrastructure and electronics foundries. Startup companies could use this approach to build electronic-photonic chips in 45nm SOI CMOS technology, for example, today, with full access to state of the art electronics on the same chip.”

The research was funded by the Defense Advanced Research Projects Agency and the National Science Foundation.

Abstracts for Optics Letters papers

We demonstrate the first (to the best of our knowledge) depletion-mode carrier-plasma optical modulator fabricated in a standard advanced complementary metal–oxide-semiconductor (CMOS) logic process (45 nm node SOI CMOS) with no process modifications. The zero-change CMOS photonics approach enables this device to be monolithically integrated into state-of-the-art microprocessors and advanced electronics. Because these processes support lateral p-n junctions but not efficient ridge waveguides, we accommodate these constraints with a new type of resonant modulator. It is based on a hybrid microring/disk cavity formed entirely in the sub-90 nm thick monocrystalline silicon transistor body layer. Electrical contact of both polarities is made along the inner radius of the multimode ring cavity via an array of silicon spokes. The spokes connect to p and n regions formed using transistor well implants, which form radially extending lateral junctions that provide index modulation. We show 5 Gbps data modulation at 1265 nm wavelength with 5.2 dB extinction ratio and an estimated 40  fJ/bit energy consumption. Broad thermal tuning is demonstrated across 3.2 THz (18 nm) with an efficiency of 291  GHz/mW. A single postprocessing step to remove the silicon handle wafer was necessary to support low-loss optical confinement in the device layer. This modulator is an important step toward monolithically integrated CMOS photonic interconnects.

We demonstrate depletion-mode carrier-plasma optical modulators fabricated in a bulk complementary metal-oxide semiconductor (CMOS), DRAM-emulation process. To the best of our knowledge, these are the first depletion-mode modulators demonstrated in polycrystalline silicon and in bulk CMOS. The modulators are based on novel optical microcavities that utilize periodic spatial interference of two guided modes to create field nulls along waveguide sidewalls. At these nulls, electrical contacts can be placed while preserving a high optical Q. These cavities enable active devices in a process with no partial silicon etch and with lateral p–n junctions. We demonstrate two device variants at 5 Gbps data modulation rate near 1610 nm wavelength. One design shows 3.1 dB modulation depth with 1.5 dB insertion loss and an estimated 160  fJ/bit energy consumption, while a more compact device achieves 4.2 dB modulation depth with 4.0 dB insertion loss and 60  fJ/bit energy consumption. These modulators represent a significant breakthrough in enabling active photonics in bulk silicon CMOS — the platform of the majority of microelectronic logic and DRAM processes — and lay the groundwork for monolithically integrated CMOS-to-DRAM photonic links.