PI’s parallel gradient search allows simultaneous optimisation of multiple degrees of freedom, elements and channels.
Scott Jordan, head of photonics, Physik Instrumente (PI), and PI fellow.
Precision positioning technology has evolved continuously over recent decades and is now becoming increasingly intelligent; highly specialised systems, which can independently determine the optimal position and orientation of individual elements in a very short time, are already available. Specific examples can be found in wafer probing and packaging in silicon photonics, as well as in nanorobotics and microassemblies. However, intelligent positioning systems will also have an impact on other applications in the future.
Towards the end of the 1970s, computers were connected by optical fibres for the first time. Since then, the use of optical transmission systems has revolutionised communications infrastructure. The positioning technology required to connect optical fibres to active optical components, such as photodetectors, waveguide structures and laser diodes, has also made rapid progress. On the one hand, mechanical manipulators (for example, hexapods) have grown ever more capable, while on the other, extremely fast, stiff piezo motors and small stick-slip positioning systems for larger travel distances with nanometre resolution have been added to established drive concepts such as DC servo, stepper and linear motors.
Thanks to reduced costs, submicron resolution linear encoders—formerly only available in very expensive positioning systems—can now be integrated into many applications, from large portal solutions to small, matchbox-sized nanopositioners. The motion controllers have also been further developed, adapted to specific requirements and extended by interfaces that include both proprietary buses and open approaches such as EtherCAT. These devices now offer highly differentiated functions, via which the individual axes and their movements can be optimised for the respective task.
Processing tedious command chains
Despite these extensive improvements in terms of the resolution, speed and compactness of the drives, as well as the expansion of controller features in terms of programming and data recording, the basic procedure has remained unchanged for a long time. Positioning systems still process long command chains, thus achieving a given position and following previously defined movement patterns. In this purely operational sense, today’s positioning systems scarcely differ from those of the 1980s.
Even back then, the increased emergence of glass fibre as a medium for information transfer created the need to align individual elements for testing and packaging. Using special software, computers were used to control the positioners and record the data from the optical power measurement systems. By aligning the photonic elements—for example, between the single-mode optical fibres (which were novel at the time) and laser diodes— predetermined sequences were repeated via successive processing of individual commands until, finally, the best possible position was reached. This limited the speed of the positioning systems and thus the throughput. In addition, it was always necessary to adjust, measure and improve the configuration accordingly until the overall result was satisfactory; inevitably, this process was time consuming.
Intelligent active alignment
Despite the fact that data transmission via optical fibres continues to conquer more and more fields of application, alternative positioning requirements need to be considered due to the development of silicon photonics.
To support the growing need for process automation for wafer probing and packaging in silicon photonics, Physik Instrumente (PI) has developed algorithms for fast, simultaneous multi-channel alignment of photonic elements and integrated these into controllers’ firmware. This significantly accelerates the respective processes because the communication between the controllers and hardware becomes considerably faster and the controllers can be tailored very precisely to the specific requirements.
The algorithms enable simultaneous optimisation of the input and output of individual elements (for example, waveguides on silicon wafers with optical fibres) and even entire arrays across all degrees of freedom. To do this, exploratory movements in the micrometre or even nanometre range are used to measure the local gradient of a given figure of merit (such as the optical power) and to automatically track this until the gradient falls close to zero. The peak, meaning the optimum position and orientation of the individual elements in respect of each other, is thus identified quickly. Another advantage of the algorithms is that they allow continuous tracking of the optimal position, thus ensuring that drift effects are countered effectively.
The parallel gradient search algorithm finds the maximum signal in less than one second. The algorithm allows tracking and compensates for drift.
The aforementioned factors mean that the processes are executed intelligently. Instead of successively executing predetermined motion sequences, the system independently seeks the fastest route to optimise the figure of merit, whereby overall optimisation of the position of all critical elements can be effectively reduced to a single step. The throughput during alignment is thus increased by two orders or more of magnitude compared with previous solutions, making the positioning process suitable for series production. As an example, in the case of double-sided alignment systems that optimise both input and output, the maximum for optimal coupling can be identified in less than a second, however the same process took several minutes to complete only a few years ago. Since alignment must be repeated multiple times during test and packaging processes, this level of throughput can greatly reduce overall production costs. These systems have proven to be a driving force for the silicon photonics business sector; indeed, they have already achieved considerable economic success and won numerous awards.
Simultaneous alignment of multiple lens elements in multiple degrees of freedom in a single step dramatically reduces production costs. Each coordinate system represents a complex manipulator.
Reduced manufacturing costs
Intelligent alignment also allows the goal to be reached quickly and cost-effectively in many other applications. In addition, a wide range of different signals can be used as a quality criterion. A good example is laser production; the mutual orientation of reflectors, gratings and other resonator components need to be optimised in a similar way, whereby the figure of merit is usually, once again, the light output. Dependencies between elements and geometric dependencies for each element are automatically taken into account by the parallelised algorithm.
Another example is camera production; multi-element lenses, of the kind that are now being installed in large numbers of smartphones, are becoming ever more complex and precise. The quality criterion might entail evaluating and optimising the modulation transfer function (MTF) or a similar metric of image quality. Once again, time-consuming process loops in this application can be reduced or eliminated via simultaneous position optimisation during alignment of the lenses and the speed of the gradient search itself.
Further examples are plentiful. The objective is always to maximise a given figure of merit for the complete system as quickly as possible via optimal positioning of the individual elements. This makes the technology extremely versatile. Diverse applications—such as camera lenses (for example, in cell phones or light detection and radar (LiDAR) optics), data storage, laser cavities and quantum computing—can benefit from this smart positioning technology. Simultaneous optimisation can replace time-consuming loops, increase throughput and significantly reduce manufacturing costs.
PI