Ushering in the Digital Age of Photonics

We have built something fundamentally new.

It is a coherent 3D image sensor chip. This chip is the solution for bringing low cost, high fidelity machine perception into the mass market. Our chip provides a low power, compact form factor imaging solution that delivers on image quality, range performance and interference immunity.

It can image the world out to 100+ meters range, with millimeter-scale range precision. It works just as well in glare from bright sunlight as it does in pitch-black darkness. This single aperture (“monostatic”), single-piece sensor is extremely robust because it contains no moving parts, even at the micro-scale, and its lack of external beam steering components like motors or MEMS mirrors enables the utmost in small, low-cost construction. In these ways, it’s unlike any time-of-flight (ToF) 3D image sensors, LIDAR sensors, or LIDAR chips in the world today.

To make this chip, we use a technology called silicon photonics, which allows us to create integrated circuitry for laser light on a semiconductor chip. However, making such a high performance chip was only possible by reinventing a new paradigm for on-chip photonic circuits: going digital.

Usually, when photonics designers want to make an on-chip optical switch, they need to supply an analog voltage, one that continuously varies, to tune the switch into exactly the right position. Imagine flipping a light switch that you needed to finesse into position, or even keep your hand on constantly to make sure it never drifted away from “on.” Now imagine controlling thousands of these switches. It quickly becomes completely unworkable, even with advanced computer control.

Instead, we created a digital photonic switch: it can be manufactured so consistently that we can control thousands as a single digital circuit. They have uniform “on” and “off” states so that they can be driven together and controlled with a very small number of digital control lines. They consume little power, and our circuitry activates only the exact switches necessary for beam steering so that no power goes to waste. This lets us make chips with 1,000 times as many switches as a chip designed with the traditional analog approach. Truly a game changer.

With this chip, we have made high performance, affordable 3D imaging a reality.

Our first proof of concept chip has 3,000 switches and is controlled using one “on” and one “off” voltage level with no calibration required. No calibration means every switch is within tolerance as-fabricated on the wafer – something photonic chips typically cannot achieve.

We don’t need a special fabrication process – we use a standard process at a high volume commercial foundry with no costly post-processing required. The system occupies a large silicon chip (~1 cm2), with switches at opposite ends of the chip remaining consistently matched with each other. That’s like someone flipping a light switch precisely in sync with another person at the other end of a football stadium.

Digital systems scale well. This is why we can carry computers in our pockets without them costing a fortune or burning up. The reason boils down to the basic building block of a digital computer: the transistor. Transistors inside a digital computer chip are switches. Each of the billions of transistors inside a computer chip has the same two discrete states, high and low (“on” and “off”, or “1” and “0”). Each transistor is not individually driven with its own arbitrary signal, meaning you can build incredibly complex systems at a reasonable cost and power.

Analog systems on the other hand, do not scale well because they operate in a continuous format rather than having discrete states. Each element of an analog system needs its own signal line, and analog systems are more susceptible to noise and other distortions so components often require extra active control loops to maintain operation. Those numerous active controls are costly and power-hungry, so running thousands is prohibitive.

Photonics systems have traditionally been built in an analog fashion. Active photonic circuits with switches, filters, resonators, or interferometers typically have an active feedback circuit dedicated to align each element’s operating wavelength. Variations in the fabrication process and temperature drift make active tuning necessary to match up components. The analog approach doesn’t scale beyond a handful of components, and pose a major barrier for building large complex photonic systems.

Our digital switch allows us to build the complex photonic system needed for creating a high performance beam steering system for our coherent 3D image sensor. Our chip contains a large dense 2D array (also called a focal plane array) of optical antennas, or pixels, much like any camera sensor. Our pixels are “active pixels” because they both transmit a laser beam and collect backscattered light. Each pixel simultaneously measures x-y-z position, velocity, and reflectance. Repeat that measurement hundreds of thousands of times per second across the full array of pixels and you image the world in real time. To accomplish this, we route transmit laser light to each pixel one by one using a large matrix of our digitally controlled photonic switches. This beam steering circuitry sits beside the pixel array, along with multiple parallel coherent receivers which perform the depth, velocity, and reflectance measurements, as well as the transmit light source and light source control circuit. Similar to a camera sensor, the full imaging solution consists of the sensor chip with a single imaging lens in front of it. And just like cameras, depending on the required performance the whole system could be designed small enough to fit into an iPhone or powerful enough to see objects hundreds of meters away.

This chip design approach gives us advantages in size, weight, power, cost and performance. Our chip allows us to easily use a large aperture lens which is vital to high SNR, long range detection, unconstrained by typical limiting factors like external mirror dimensions. Our chip can also run in a single-aperture, or monostatic, configuration in which the same lens and pixel array is used for both transmit and receive. This is advantageous over the more-common bistatic configuration where separate transmit and receive apertures increase system size, weight and power, and require expensive precision optical alignment. The high density integration of on-chip photonics enables parallelization of multiple transmit/receive paths without interference, which multiplies our data rate without significantly increasing cost.

The fact that all the elements of the image sensor are on a single chip is notably distinct from other so-called “LIDARs-on-a-chip” in which most of the LIDAR system is on chip but one or more of the key LIDAR system building blocks are not integrated on chip. Beam steering is the most challenging function to integrate on chip, and none of the “LIDAR chips” on the market today have this. With our approach, this part is not only achievable but straightforward and robust.

Preorders for the Voyant Blue product line, which uses this technology, are available now.

Steve Miller

AUTHOR

Share This Post