For researchers who study the brain, it’s important to be able to keep up with events happening quickly, and over a large field of view. Neuronal circuits can cover more than a cubic millimeter in volume, and communication happens at the millisecond timescale. Liquid crystal on silicon (LCOS) spatial light modulators (SLMs) are the technology of choice for stimulating neurons with targeted laser illumination. Our collaborators in the Deisseroth Lab at Stanford came to us asking for a larger, faster, SLM for cutting-edge holographic photostimulation.
To meet this challenge, we designed and delivered the MacroSLM, with a 1536×1536 square pixel array to maximize addressable field of view, and high-voltage LC drive to maximize speed. Particular features of the SLM include custom FPGA board for handling high data rates, large pixel size for minimizing pixel crosstalk, and temperature control to handle heating effects from the high-voltage controls and high-power laser illumination. We also designed an FPGA implementation of the overdrive method for increasing liquid crystal switching speed, allowing us to overcome the significant data bottlenecks that limit frame rates for large arrays. Using this system, we demonstrated efficient 500 Hz hologram-to-hologram speed at 1064 nm operating wavelength.
With these cutting-edge speeds and array sizes, our customers were able to publish groundbreaking science.
- Neural Mapping
Our recent SLM developments have been driven by the needs of neuroscientists, who use the technology for holographic photostimulation in optogenetics. In photostimulation, engineered light-sensitive neurons are induced to fire when illuminated, allowing researchers to controllably activate neuronal circuits. Because neuronal circuits are interlaced with one another, selectively activating a particular network requires illuminating collections of individual cells rather than large areas.
The problem of beam steering for photostimulation is a particularly challenging one, since neuronal circuits can involve many hundreds of neurons, can cover more than cubic millimeter in volume, and have dynamics on the millisecond timescale. LCOS SLMs are the technology of choice for 3D holographic beam steering, with high-resolution arrays of phase-modulating pixels that allow hundreds of beams to be independently steered and focused.
Most of the LCOS SLMs currently available are repurposed commercial display devices whose speeds are limited to around 60 Hz, much slower than the kilohertz timescales of neuronal circuits. A faster speed response is particularly important for closed-loop experiments, in which the displayed hologram is updated in real-time response to the dynamics of the neuronal circuit. This requires not only the liquid crystal (LC) pixels, but also the entire hologram data pipeline, to be designed to be capable of responding to triggers at these speeds. Additional design considerations such as pixel size, pixel count, and pixel crosstalk help determine the field of view that can be efficiently addressed with holographic beam steering. Power handling is also an important consideration, since the near infrared (NIR) femtosecond sources used in optogenetics experiments can have average powers of multiple Watts, which can cause enough heating to affect the calibration of the LC response.
Photograph of the MacroSLM, along with its op-amp board and driver board.
For our design, we chose a square array size of 1536×1536 pixels, as a compromise between fabrication cost and data handling. Although most LCOS SLM arrays available are rectangular, this is a reflection of their being repurposed display devices. In practice, SLMs in holographic beam steering are usually illuminated by circular beams, so the most relevant pixel count is that of the shortest dimension. A square array takes maximum advantage of available backplane space and avoids slowing frame rates by loading voltages onto unused pixels.
Although a larger pixel pitch corresponds to smaller diffraction angles, the field of view is not affected as long as the magnification of the optical system is designed to image the SLM to the size of the pupil of the objective lens. For the MacroSLM we choose a relatively large 20 µm pixel pitch. This resulted in larger fill factor, lower pixel crosstalk, and higher phase stability due to larger pixel capacitance.
The combination of a large pixel pitch and high pixel count leads to an active array area of 30.7 × 30.7 mm. Although it necessitates larger optics in the system, it requires less magnification change, and therefore fewer of the associated aberrations, when used in combination with the large back apertures of the low-magnification high-NA microscope objectives that are increasingly common in photostimulation. A large array also increases the SLM’s capabilities for power handling.
For our SLM we choose high-voltage (0 – 12 V analog) pixel addressing for fast LC response and to enable strong overdriving of the pixels to further speed up LC switching.
We used backplane Peltier heating/cooling to account for the effects of self-heating and high-power illumination. With active control, we can maintain the LC temperature at a level where the available phase stroke remains high, yet LC viscosity is low, resulting in faster and more controllable phase response.
Data handling is another significant part of our speed effort, since the system must be capable of calculating the required transient overdrive voltages to achieve fast LC switching from phase to phase at each pixel, while loading the transient 1536×1536 images onto the SLM pixels at ~1250 Hz continuous frame rate. We use a custom FPGA solution for handling these high data rates, including on-board storage of 2045 images, on-board application of spatially-varying voltage calibrations, and on-board calculation of individual transient voltages for every pixel. The transient voltages must be calculated on-the-fly to allow for flexible switching between stored target holograms; otherwise, the stored images could only be displayed in a single predetermined order.
With this data-handling pipeline, the ultimate limitation on our hologram display frame rate is the response time of the LC, rather than the response of the electronics. As a result, our system can be triggered at arbitrary rates exceeding 1 kHz.
We characterize the speed of this SLM as the triggered rate at which the system can switch between typical spot-forming GS (Gerchberg-Saxton) holograms while maintaining >90% of its steady-state efficiency. This provides a more complete and relevant picture than examining rise time or fall time alone, or examining the speed of a single phase transition, since it takes into account the entire pipeline of trigger response, data management, and pixel addressing as well as LC response over a range of phase transitions. In addition, we make sure to measure speed at 1064nm, in the wavelength range where we plan to operate, since LC response is often more than 3x faster at visible wavelengths.
Hologram-to-hologram switching of the BNS 1536 SLM
Operating wavelength is 1064 nm (LC response at visible wavelengths is typically >3x faster). Test holograms are 8-spot Gerchberg-Saxton (GS) spot-steering holograms applied to the SLM 1536 x 1536 array. One hologram steers a spot to a single target location where we place our photodetector, plus 7 other random locations. Another set of 10 holograms steers spots to 8 random locations with the photodetector’s location excluded. In this way, we sample a representative variety of phase transitions and complex holograms. Transient voltage (“overdrive”) frames are calculated on-the-fly by on-board FPGAs and delivered at ~1250 Hz. Blue: System triggered at 100 Hz, indicating the diffracted signal strength at moderate speeds.
A) Red: System triggered at 500 Hz. Peak-to-peak signal amplitude is 96% of the slow-speed amplitude.
B) Red: System triggered at 600 Hz. Peak-to-peak signal amplitude is 88% of the slow-speed amplitude.
A data-handling system that enables interruptible image downloads is a key requirement for this. Otherwise, if triggers can only be handled whenever an image download is complete, this restricts the trigger interval to integer multiples of the SLM’s base refresh rate. Latency and jitter can both be significant, especially for SLMs with low refresh rates. An SLM with a 60 Hz refresh rate could have a trigger response latency that varies from 0 to 17 ms depending on when the trigger arrives within the refresh cycle.
With the ability to interrupt the download of an image onto the SLM’s pixels, the latency between a trigger arriving and the voltage changing on our SLM is 6 µs with a range of +/-3 µs, so that the transition to a new hologram can be very predictably initiated. This latency is well within the LC response time and is therefore insignificant under photostimulation experimental conditions.
Triggering Hologram Transitions at Arbitrary Intervals
A) Trigger signal (SLM triggers on falling edge) with frequency sweeping from 500 Hz to 100 Hz at a 10 Hz period.
B) SLM optical response. Test holograms are 8-spot Gerchberg-Saxton (GS) spot-steering holograms applied to the SLM 1536 x 1536 array. One hologram steers a spot to a single target location where we place our photodetector, plus 7 other random locations. Another set of 10 holograms steers spots to 8 random locations with the photodetector’s location excluded.
Science, vol. 365, no. 6453, Aug. 2019, doi: 10.1126/science.aaw5202
Optical Trapping and Optical Micromanipulation XVI, San Diego, United States, Sep. 2019, p. 3, doi: 10.1117/12.2528558