Scott Jordan, Business Developer – Photonics and Nano Automation Technologies and Stefan Vorndran, Vice President Marketing – North America, Physik Instrumente L.P. (PI)
Latest advancements in algorithms and motion systems.
Why Photonics Applications Are Expanding
The world’s appetite for data continues to grow at a steady pace, with new applications emerging every week. Even during the height of the COVID-19 pandemic in 2020, data consumption grew by more than 20 percent, making it vital for information to be efficiently routed and conveyed.
The photonics industry is growing alongside, as its technologies are found in many mainstream applications. These include various light detection and ranging (LiDAR) solutions and sensors in wearables and vehicle cameras, as well as functions in quantum computing, high speed cables, and even photonic computing where photons, rather than electrons, are used for logic. Silicon photonics (SiPh) is the technology behind these innovations, which combine optics and electronics on the same wafer to deliver significant performance gains in a smaller final product.
Based on the expanding and diversifying applications of SiPh, especially with the emergence of consumer applications, production volumes need to increase 1,000-fold in the next few years to keep up with the demand. This paper discusses the latest advancements in alignment automation that will help to make this possible.
The Difference Between Optics and Electronics
All optical circuits first need to be tested, starting with the wafer and continuing through multiple stages of assembly. This requires much greater physical accuracy than probing electrical circuits, often two to three orders of magnitude higher, as well as additional degrees of freedom (DOF).
While probes can easily reach electrical test points, since pad sizes of approximately 30 microns are practical targets, SiPh structures need accuracy between 20 and 50 nanometers, calling for novel technologies with greater precision. This precision cannot be resolved with cameras or microscopes, and alignment must be ‘active’, meaning the positioning must be based on optimizing the optical power.
The Challenge of Upscaling SiPh Production
The production of photonic integrated circuits involves particularly demanding nanopositioning tasks, and vision- or fixturing-based approaches are inadequate, especially as variations exist between devices. The need for repetitive alignment for all manufacturing stages, from initial probing and grading on the wafer through final assembly and packaging, including multiple steps of placing and bonding functional elements interspersed with more testing, has long been regarded as an obstacle for scaling up production, as it accounts for up to 80 percent of the total costs. In fact, the traditionally high cost of active alignment methods has been one of the main drivers of the multi-decade search for passive alignment technologies. Nevertheless, precision active alignment is now possible, eliminating many of the associated costs and enabling a cascade of savings across the manufacturing workflow.
Figure 1: Testing and packaging today’s photonic devices can be a huge challengeacross multiple DOF. The alignment of multi-channel devices, such as fiber-opticalarrays, used to be a slow, repetitive process before modern parallel algorithmswere developed. (Image courtesy PI)
Passive or Active Alignment?
In non-photonics manufacturing, precision alignment can be achieved using either an active or passive strategy. An example of the latter is a dovetail joining two boards at right angles to each other, where the edges perfectly align. However, passive positioning is often not practical for applications requiring higher precision, such as the alignment of optical fibers to chips or other components, because the tolerances are too tight, which makes the manufacturing processes unreliable.
Furthermore, as the design becomes more technical, the machine tools and processes that can meet the required tolerances grow more expensive and, inevitably, the number of rejects rises, resulting in additional costs. Besides, there are some device tolerances that cannot be addressed with better fabrication techniques, such as fiber-core centration, and machining the world's finest V-groove is of little value when device-to-device variability exceeds the necessary alignment tolerances.
Nevertheless, clever passive alignment solutions have emerged that show great promise for specific applications, such as the Photonic-Plug (Teramount), but the need for active alignment is not going to disappear.
Comparatively, active alignment maximizes performance through the use of robots to automatically align devices and, given the cost savings, flexibility and speed described below, it will remain the preferred choice for photonic device manufacturing in the future. Active alignment works by optimizing coupling performance, and the latest innovation described in this paper performs this cost-effectively across numerous channels in multiple DOF, which is required as SiPh has the ability to print several circuits with multiple inputs and outputs on a single chip.
Active Alignment: What is Necessary?
Active alignment requires intelligent control electronics and high precision mechanics. It involves autonomously controlling the mutual position of the photonic device’s components during the manufacturing process.
Traditionally, this has been a time-consuming process, because optimization of several channels, inputs and outputs in multiple DOF, required numerous iterations until the result was right. Using legacy approaches, this time will only increase as devices grow smaller and more complicated, increasing the cost further. For example, alignment in a multi-lens assembly is far more demanding than for a single element, as is a SiPh chip with multiple channels instead of the single fibers people were attaching in the late 1990s. Fortunately, breakthrough control algorithms can now perform all the necessary sub-alignments across multiple DOF, channels and operating systems in parallel, accomplishing the whole job in one quick step, while eliminating repetitive looping iterations and reducing alignment time by typically 99 percent.
Precision Positioning of Signal-Carrying Elements
The central challenge in active alignment is accurately positioning signal-carrying elements, including single fibers or fiber arrays, lenses, interposers or other chips, with respect to coupling points such as grating couplers or edge facets. These couplings can be into or out of passive and active photonic structures, for example, waveguides, vertical cavity surface emitting lasers, or photodiodes. Typically, alignment accuracy down to tens of nanometers is mandatory when coupling photonic components, to generate the highest possible optical power transmission and lowest attenuation.
Precision Automation: Finding the Main Peak
In almost all photonic component alignments, the signal, for example coupled optical power, rises to a global peak when optimally aligned and falls when out of position. This is generally a universal pattern, and often false lower or local peaks can be seen, which must not be selected for optimization. Finding that global peak, however, has traditionally been a time-consuming task, especially when it depends on several DOF.
Power distribution of an optical component shows multiple peaks. Modernalgorithms and precision mechanics can help determine the main mode andfind the exact peak quickly. (Image courtesy PI)
Advanced Algorithms and Fast Mechanics
A novel combination of advanced algorithms and rapid mechanisms, for example, the award-winning Fast Multichannel Photonics Alignment solutions (FMPA) from PI have largely accelerated throughput. FMPA simultaneously optimizes coupling between photonic elements across channels, inputs and outputs, and DOF, even when these independent variables influence each other, through novel functionalities that include rapid first light detection, a vibrationless area scan, and an innovative parallel gradient search that can perform real-time, multivariate tracking optimization.
Coupling in previous approaches had to be optimized in a sequential, looping fashion, constantly going back and forth and readjusting at each axis to gradually reach a consensus. This took a considerable amount of time, often several minutes, which had a significant impact on costs, efficiency and scalability.
Comparatively, FMPA can achieve all of these adjustments in just one single step with a simple set of commands, which can be performed in just seconds. Furthermore, optimization time is largely independent of the number of alignments being performed, since they are performed in parallel. This significant reduction in alignment time, typically two orders of magnitude, massively reduces costs, making processes such as probing SiPh wafers economically viable.
Alignment speed is crucial in reducing production costs. (Image courtesy PI)
Active Alignment: Different Processes
There are two fundamental types of alignment processes for photonic components: Area scans and gradient searches. Area scans determine the peak of a measured figure of merit, such as optical power, modulation transfer function (MTF) or modal purity, within a defined region, and can be used to precisely characterize the optical fingerprint of a component, as well as separate global and local maxima. Gradient searches provide fast final optimization and tracking, and FMPA’s unique implementation can optimize one or more couplings in multiple DOF at once, as well as track them to mitigate drift processes, disturbances, etc. These processes are discussed in more detail below.
Area Scans
Area scans, which scan an area to determine the approximate location of the highest coupling peak, are used for a variety of tasks, including:
▪ Detecting first light;
▪ Characterizing (profiling) a coupling, which is important for process control;
▪ Localizing the main mode of a coupling for subsequent optimization by a gradient search. This sequence forms a powerful, hybrid approach that helps to prevent locking onto a local maximum. The area scans built into FMPA include unique single frequency sinusoid and spiral scans. These are much faster than raster or serpentine scans, as they are truly continuous and avoid settling the requirements of the stop-and-start motions used in traditional scans. Additionally, the frequency is selectable to avoid exciting structural resonances. It is also possible to select a constant velocity spiral scan, which allows the acquisition of data with constant density across the spiral.
Figure 4a: Sinusoidal area scan for detecting first light. (Image courtesy PI)
Figure 4b: Spiral scan using a hexapod/piezo approach. A fine and coarse scancan be executed simultaneously. (Image courtesy PI)
Gradient Searches
Gradient searches perform a small, approximately circular dither motion between devices to modulate the coupling. This variation of the figure of merit allows the local gradient of the coupling to be determined in real time. The controller automatically observes modulation and drives the alignment to reduce the modulation until it falls to zero, indicating full optimization.
Figure 5: Graphical depiction of gradient determination via a circular dither, whichmodulates the observed coupled power (or other parameters). The modulationphase with respect to the dither indicates the direction towards maximum whileits amplitude falls to zero when optimized. (Image courtesy PI)
|ε(Θ)| = Δ I = (Imin−Imax)/Imin
The observed gradient can be expressed by the above equation and
serves as a measure of alignment error. (Image courtesy PI)
From the observed modulation, the controller can automatically deduce the local gradient via a simple calculation (Figure 6). Note, that the gradient (Δ I) falls to zero when optimized. Any axes in an FMPA system can perform these types of alignments, subject to the physical capabilities of the axes.
Gradient searches are rapid and precise tools for transverse optimization, but they can also be used in other functions, including:
▪ In a single linear axis, which is ideal for localizing the beam waist in a lens coupling;
▪ In a gimbaling or swivelling fashion to optimize an angular orientation; or
▪ In a rotation about one channel axis of a multichannel device, to bring all elements of the device’s arrayed input/output (I/O) channels into correspondence.
These algorithms are suitable for all kinds of optimizations, including bulk optic, cavity and pinhole alignments.
A graphical depiction of a digital gradient search (hill-climb algorithm).Once the main mode has been identified through an area scan, the algorithm will reach the peak quickly. The main peak is automaticallyfine-tuned in all DOF theapplication requires. All of this is achieved in parallel with one command by theuser. With today’s photonic devices,it cannot be assumed that the peak is symmetrical,follows a Gaussian distribution, or has a circular cross-section. Furthermore,both edge and diffractive-coupler scenarios must be met. (Image courtesy PI)
Compensating for Position Changes
In packaging, photonic devices must be permanently bonded together and connected to fibers to be able to transmit signals, which is typically done with a UV-curing epoxy resin. However, polymerization that occurs during the curing process increases stress and causes displacement of the components. For this reason, a tracking mode within the FMPA algorithm is useful for compensating positional changes of the optical elements in real time, in the early stages of polymerization, making the necessary adjustments, and fine-tuning the alignment to maintain optimum positioning.
Solving the First Light Problem
Before the aforementioned optimization processes can even start, an optical signal, above the noise level, needs to be detectable. This process is called first light detection. Finding first light has been a time-consuming procedure in all industrial photonics alignment applications, including wafer probing and device packaging. It is particularly arduous in devices with inputs and outputs where both sides must be aligned to achieve even a threshold level of coupling.
Traditional First Light Search Algorithms
Signal finding is typically based on performing cyclic patterns such as Archimedean spirals or sinusoidal raster scans at the micron to submicron scale (see Figures 4a and 4b). In the case of large device-to-device variations or indeterminate fixturing, these repetitive, tightly spaced scans can take a significant amount of time to complete depending on factors such as the area to be searched, whether inputs and outputs need to be simultaneously aligned, and so on.
New First Light Capture Process
A breakthrough has been achieved in the form of a novel, embedded search and alignment algorithm (patent pending), that promises to revolutionize the field. The algorithm, called PILightning, runs embedded on PI’s advanced controllers. It enables highly dynamic mechanics, such as piezo scanners or direct-drive air bearing stages, to achieve significant economic gains in production over previous first light search algorithms. This new process is fully automated and virtually instantaneous, eliminating the need for extensive calibration or manual intervention.
First light detection with PILightning. (Image courtesy PI)
PILightning is based on a new search method with integrated AI-based real-time executive function (see Figure 8). It dramatically reduces the time required to find first light in single-sided and double-sided couplings and in loopback (omega) waveguide configurations. Once first light is detected, the FMPA fast gradient search algorithm takes over, utilizing real-time feedback control to rapidly optimize the alignment in parallel across degrees of freedom and channels. Depending on the application, a tracking algorithm can also be activated to maintain maximum coupling efficiency – important, for example, in curing situations.
Orders of Magnitude Improvement
Tests have shown that PILightning typically reduces first light capture by one order of magnitude or more in single-sided alignment applications. Gains of more than two orders of magnitude are achieved in double-sided applications. The larger the search area and (as with the FMPA parallel optimization functionality) the more complex the alignment, the more significant the gain.
Mechanisms for Different Requirements
While the algorithms mentioned above apply to all kinds of mechanical alignment hardware, various demands, such as size, travel range, DOF, etc. on alignment mechanisms require different configurations and technologies.
Optical Power Meters: Bandwidth and Dynamic Range Rule
A highly sensitive power meter with wide dynamic range and high bandwidth is also crucial to support the optimal alignment of SiPh components. Typically, 20 kHz signal bandwidth and a dynamic range of six orders of magnitude with logarithmic output are leveraged to make full use of the FMPA algorithms.
High speed optical alignments require optical power meters with high
Gantry Platforms for Large Format Applications
Processing densely packed nanoscale structures over large areas for various applications, including semiconductors and SiPh testing and manufacturing across circuit boards, trays, carriers and other large substrates, requires high precision, high throughput solutions delivered on reliable platforms. Novel gantry systems are suited perfectly for these applications, helping to significantly improve throughput and deliver repeatable and robust results in a compact form factor.
Some modern gantry systems, such as those provided by PI, use power-dense ironless linear motors to accommodate the high dynamics required in photonics applications. These efficient direct drive motors can have two high resolution linear encoders, with nanometer or sub-nanometer resolutions, on each gantry axis, which deliver repeatable positioning even in the most demanding duty cycles. Mechanical or air bearings are available, with the latter providing multiple advantages, including a cleaner operation with no risk of contamination that could damage the delicate optical circuits. Controllers with and without built-in FMPA functionality are available.
Mini Gantry. (Image courtesy PI)
Hexapod Mechanisms: Parallel Kinematics
Hexapods are extremely versatile motion and positioning systems, which are based on six actuators placed between two plates operating in parallel. They provide all six DOF in a very compact and stiff package, making them an obvious choice for photonic integrated circuit (PIC) test and packaging applications. These require at least four DOF of alignment automation with a flexible variable rotational pivot point, and often benefit from five or six axis optimizations. Additional power and flexibility come from mixing and matching different motion technologies. For example, the addition of a 3-axis piezo flexure scanner offers unmatched nanoscale resolution and accelerates the alignment process due to the low inertia of the piezo mechanism, which leads to a high scanning frequency to increase productivity and generate a tight, responsive lock-on.
PI hexapod controllers also offer a programmable coordinate system, allowing them to be mounted at any angle while maintaining the desired XY plane. This also provides a flexible definable pivot point that allows the user to specify the center of rotation anywhere inside or outside the hexapod envelope. The definable pivot point is invaluable when rotation around a fiber tip or beam waist is required.
Hexapod System. (Image courtesy PI)
Modular Alignment Systems: Serial Kinematics
Modular alignment systems that consist of a combination of individual stages, known as serial kinematics, are guided by one of two designs: mechanical or air bearings. Air bearings are preferred as they provide precise, high dynamic motion that is free of friction, wear, and maintenance. Furthermore, air bearings do not generate particles as they neither contain lubricants nor are subject to abrasion caused by friction.
PINovaLign42 Photonics Alignment System in a double-sidedarray alignment application. (Image courtesy PI)
Mechanical Bearing Stages
Mechanical bearing stages are cost-effective and rigid solutions that do not require a clean air supply. They have crossed-roller bearings that are generally the preferred choice due to their high stiffness and precision. A multitude of configurations is available, from compact voice-coil-driven systems providing 7mm XYZ-travel to larger stages covering 250mm or more. Direct drive motors, such as 3-phase motors, and voice coils are normally used, which help to deliver the highest performance and longest lifetime.
ACS Motion Control and Drive Modules
The photonics alignment systems pictured in Figures 11 and 12 operate with ACS Motion Controllers with integrated alignment firmware, as well as with system metrology that measures repeatability and accuracy with error mapping.
Additional Alignment Applications Outside of SiPh
In addition to SiPh and fiber optics, precision alignment can increase yield and throughput in other applications with photonics-related components. In laser manufacturing, for example, bulk optics, cavities and diffractive elements need to be aligned to optimize the output beam's characteristics.
This can be a time-consuming process, but automation can significantly reduce manufacturing time, and increase productivity and yield.
Similarly, high speed alignment can help manufacturers keep up with the ever-increasing demand for high resolution cameras which are, for example, used in smartphones and vehicle cameras. FMPA is a novel technology that drives this precision and speed, provides nanoscale autonomy to microrobots, and optimizes the position of multiple optical elements. With more sophisticated optical devices being produced each year, FMPA applications will continue to expand, ensuring precision positioning stays at pace with the rapidly evolving computing, imaging and photonics industries.
Alignment of multiple lens elements is a complex task that can be accelerated by algorithms which are built into the controller firmware, along with novelmechanical alignment units, allowing the parallel alignment of multiple lenses in a single operation. Typically, the position of a lens system in front of a sensor is optimized usinga calculation such as framewise MTF, which provides information about the sharpness of an image, with a better value signifying a better alignment. In other applications, suchas single-photon quantum emitter studies, photon count is another signal that can be used to assess alignment. Basically, any sufficiently fast figure of merit that risesto a peak and then falls as a function of position can be optimized efficiently with FMPA. (Image courtesy PI)
Summary
The process of aligning components is essential at every stage of photonic device production, starting with initial testing and grading on the wafer. This is traditionally a time-consuming and expensive task, and aligning hundreds or thousands of components on each wafer can be a daunting and, often, cumbersome task.
Traditional methods that take several minutes, such as aligning the vertical-cavity surface-emitting laser and photodiode channels on both ends of a photonic cable, may be valid for low volume, specialized items, but as demand exponentially increases, they become unfeasible. This is when automation becomes essential. Technologies such as FMPA clearly address this need, and are capable of meeting the current and future demands of the industry.