Over the past several decades, silicon has undoubtedly been the crown jewel of the semiconductor industry’s transformation. But as Moore’s Law plateaus, the increasing complexity of circuits, and the exponential growth of data-intensive applications, companies need more innovative ways to compute, store, and transmit data faster. As a result, size, speed, and power have become essential forces for handling both advanced intelligence and computing needs.
Silicon photonics has already earned a stronghold for its impressive performance, energy efficiency and reliability compared to conventional electronic integrated circuits. The overall speed requirements are now fast enough, which takes advantage of the strengths of the technology to move data efficiently over ever-shortening distances. Meanwhile, artificial intelligence (AI) is pushing computing to a point where electronic components need to communicate across distances to integrate and combine multiple XPUs (application-specific processing units).
Research and commercialization of silicon photonics has seen a parallel boom, with markets such as data and communications applications, optical computing, and high-performance sensing applications such as LiDAR also seeing their advantages come to life. According to research by LightCounting, the market for silicon photovoltaic products is expected to increase from 14% in 2018-2019 to 45% by 2025, indicating an inflection point for technology adoption.1
This comes as no surprise as more companies are collaborating and investing in silicon photonics to solve current I/O and bandwidth bottlenecks, along with the challenges of existing discrete components to achieve accelerated growth and performance.
This shift in market ambition did not happen overnight.
How We Got Here: From Evacuated Tubes to Fittings
From the 1920s to the 1950s, all electronic components were discrete items—primarily vacuum tubes that controlled the flow of electrical current between the electrodes to which a voltage would be applied. Soon after, the first transistor was invented, marking the beginning of the extraordinary progress of the electronics industry. The industry then expanded further with the advent of integrated circuits – a single chip containing millions or billions of transistors integrated. The development of microprocessors soon followed, benefiting everything from pocket-sized calculators to household appliances.
Classical microprocessors advanced in speed through the 1990s, but since about 2003, mainstream processors have hit the 3GHz clock wall. Despite the increase in the number of transistors, not only did the processors overheat, but even smaller transistors ceased to be more efficient. This means that the transfer of data from a computer chip to a memory or other computing chip via copper wire is no longer sustainable, no matter how short the distance, and this has increased various degrees of difficulty.
The light at the end of the tunnel has become silicon photonics.
The industry is beginning to see the promise of harnessing the power of light and combining semiconductor lasers with integrated circuits. The rich history and development of electronics has inspired researchers and engineers to find new ways to integrate functions on a chip and to use light beams with well-defined wavelengths to be faster than electrical connections.
Today, a similar physical path occurs with chip electrical interconnects at 100Gb/s per lane (four levels at 50Gb/s), where a significant amount of tie power must be added to push the signal over the copper wire. In fact, at 200Gb/s per lane (four levels at 100Gb/s), this problem gets even worse.
On the other hand, optical interconnects do not suffer from the same problem because fibers can easily transmit several terabytes of data. Simply put, utilizing photonics to transmit information offers significant improvements in speed and energy efficiency compared to electronic approaches.
Race for strength and speed
Every bit of acceleration comes at the cost of consuming more energy. As circuit designs and complexity grow — whether it’s high lane counts, dense sensing, or terabit connections — teams will inevitably need to move away from separate roads. We’re already seeing this shift within the industry, with companies moving from discrete elements to silicon photonics, and eventually to platforms with on-chip lasers for additional optical gain.
In the world of interconnection, there is still a lot of focus on data rate per pin. Today, 100Gb/s interconnection is done at four levels with 50Gb/s to get twice the amount of data going through a 50Gb/s data link. But a 200Gb/s connection ends up pushing more power through it to get that signal over an electrical hookup. Eventually, the amount of energy consumed becomes an issue, especially when pushed over greater distances. Thus, teams cannot ingest any further data through these electrical connections.
This is not the case with optical fibers. Think of fiber optics as an open highway of a thousand lanes. A compute box can be designed to be the size of a data center without sacrificing going for smaller interconnection scales. But when using separate component parts, the size of the processors is limited by their interconnection.
Today, some companies are taking a 12-inch chip and making one huge chip out of it, with interconnects designed to keep all the cores running at high speeds so the transistors can work together as one. However, as modern computing architectures approach their theoretical performance limits, these bandwidth requirements increase in complexity and size, making laser integration more expensive. With standard silicon photonics, one would need to connect the lasers separately, which does not lend itself well to multiple channels.
Integrated Laser: A engineered match for next-generation designs
Laser integration has long been a challenge in silicon photonics. The main areas of concern refer to the fundamentals of physics at the design level and the incremental cost associated with fabricating, assembling, adding and aligning discrete lasers to the chip. This becomes an even bigger test when dealing with the increased number of laser channels and overall bandwidth.
So far, silicon photonics has seen many photonic components embedded in a chip, but the key component missing so far is the integrated gain. Gain-on-chip departs from standard silicon photonics to achieve a new level of integration and enhance computational capabilities and overall processing. This helps provide high-speed data transfers between and within the chips in far greater numbers than can be achieved with separate devices. The technology’s advanced ability to drive higher performance at lower power or reduce the cost of design and manufacturing processes has helped drive its adoption.
Take ultrasensitive sensing applications such as LiDAR. For coherent LiDAR, the light from the transmitter must be mixed with the receiver to back out the information, which is why it gets better range information with less power. With an integrated laser on a single chip, this process becomes easier because you can separate the light and place it on a different part of the circuit. If you were to do it with separate components it would require quite a bit of packaging. While the extent of their benefits depends on the complexity of the circuit, this is the primary reason why approaches such as continuous coherent wave (FMCW) LiDAR can benefit from an integrated approach. Combined optics and system-on-a-chip (SoC) interfaces (Courtesy of OpenLight)
Does silicon optics replace electrical connectors?
Processing materials such as indium phosphide for semiconductor lasers directly on the photonic silicon wafer manufacturing process reduce cost, improve energy efficiency and wafer acquisition, as well as simplify packaging. With monolithically integrated lasers, productivity remains high, while scaling the design with discrete components leads to unacceptable throughput. At this point, even dozens of components in a circuit are revolutionary.
However, just like the adoption of any new technology, the ecosystem goes through a learning curve. Most manufacturing units are still familiar with bonding materials such as indium phosphide and gallium arsenide (used in the laser industry) to silicon. Due to their different physical and thermal properties, some barriers to entry pertain to discrete approaches that need to be overcome. In short, the fab that has spent decades nailing 8- or 10-inch chips and the purest of various materials now needs to learn how to use newer materials and a different design space that makes the process unique.
Integrated gain silicon photonics
With the pace at which silicon photonics technology is increasing, companies and foundries will inevitably expand collaboration and R&D investments to enable a robust photonics ecosystem of integrated components and solutions. Because transceivers can accommodate eight or 16 lanes, silicon photonics will be the only technology that can deliver the required performance at lower power and at a reasonable cost.
Some would argue that with the varying complexity of each application and the circuitry being at its core, there may still be some unknowns as to its potential in areas such as full autonomy or Advanced Driver Assistance Systems (ADAS), but no way in which its benefits will be unseen. At some point, silicon photonics will mature enough that some key metrics including bandwidth, cost, and power per bit will be sufficient to replace electronics. From now on, the main value of switching to optics will be its reach.
1. See www.lightwaveonline.com/14177636.