Texas Instruments Phase Light Modulator (PLM)

• Texas Instruments PLM is a MEMS-based phase-only spatial light modulator featuring a 1.09 MP micromirror array and 4-bit phase control for precise adaptive optics applications.
• It employs interferometric calibration and lookup table methods to accurately translate discrete mirror motions into effective phase shifts.
• Demonstrated in adaptive optics and beam tracking, the PLM achieves kilohertz update rates and over 80% diffraction efficiency in practical wavefront shaping.

The Texas Instruments Phase Light Modulator (PLM) is a MEMS-based, phase-only spatial light modulator built on Texas Instruments’ Digital Light Processing platform. In the reported implementations, it consists of a 2D mega-pixel array of electrostatically actuated piston-type micromirrors whose vertical height can be independently adjusted with 4-bit precision, enabling direct phase control at each pixel. Within recent adaptive-optics and wavefront-shaping work, the PLM is presented as a low-SWaP, chip-scale solution for simultaneous wavefront correction and beam tracking in real time, featuring over one million actuators and operation at kilohertz update rates (Rocha et al., 2024, Bass et al., 7 Sep 2025).

1. Hardware architecture and physical characteristics

The smallest commercial device described in the wavefront-correction study is the 0.67″ PLM, hosting a $1358 \times 800$ array of piston-type micromirrors, or approximately $1.09$ million actuators. Each mirror is a square of side length $10.8\,\mu\mathrm{m}$. A larger-format 0.98″ PLM increases both resolution and speed, while retaining the same underlying architecture as an electrostatically actuated micromirror array (Bass et al., 7 Sep 2025).

In the engineering characterization of the DLP6750 EVM, the array is likewise specified as $1358 \times 800$, with pixel pitch $10.8\,\mu\mathrm{m}$, fill factor approximately $95\%$, and mirror-coating reflectivity greater than $95\%$. Beneath each mirror are four interleaved, independently driven electrodes; by applying the appropriate binary on/off pattern across these four electrodes, the mirror is electrostatically pulled to one of 16 discrete vertical positions (Rocha et al., 2024).

Parameter	Reported value
Mirror array format	$1358 \times 800$ (1.09 MP)
Pixel pitch	$10.8\,\mu\mathrm{m}$
Fill factor	$\approx 95\%$
Phase levels	16 (4 bit)
Mirror stroke	$1.09$0
Continuous update rate (EVM)	$1.09$1
Active aperture	0.67″ diagonal ($1.09$2)
Package size	$1.09$3

The operating principle is electrostatic piston motion: each pixel is a rigid mirror pulled toward a backplane, producing a locally controlled phase delay. Because the actuation is capacitive, static power draw is essentially zero and only the digital driver incurs dynamic switching losses. The SWaP-oriented description reports total driver electronics on the order of a few watts, package size $1.09$4, and chip weight less than $1.09$5; the system-level comparison for the PLM solution reports less than $1.09$6, less than $1.09$7, and less than $1.09$8 (Bass et al., 7 Sep 2025).

2. Phase modulation, quantization, and calibration

For reflective operation, a vertical displacement $1.09$9 of a mirror increases the round-trip optical path by $10.8\,\mu\mathrm{m}$0, so the phase shift imparted to a beam of vacuum wavelength $10.8\,\mu\mathrm{m}$1 is

$10.8\,\mu\mathrm{m}$2

Each 4-bit code $10.8\,\mu\mathrm{m}$3 corresponds to a discrete height $10.8\,\mu\mathrm{m}$4, yielding $10.8\,\mu\mathrm{m}$5. In the DLP6750 EVM characterization, the total piston range is approximately $10.8\,\mu\mathrm{m}$6–$10.8\,\mu\mathrm{m}$7 at maximum mirror-bias voltage, sufficient to deliver a $10.8\,\mu\mathrm{m}$8–$10.8\,\mu\mathrm{m}$9 phase shift at design wavelengths between $1358 \times 800$0 and $1358 \times 800$1. In the low-SWaP wavefront-correction report, four-bit discrete levels are stated to cover at least $1358 \times 800$2 at $1358 \times 800$3, with 4-bit depth yielding a steering precision less than $1358 \times 800$4 (Rocha et al., 2024, Bass et al., 7 Sep 2025).

The finite 16-level quantization introduces a maximum quantization error of roughly half the step size, $1358 \times 800$5 rad at mid-range. In practical hologram generation, this quantization is reported to yield diffraction efficiencies on the order of $1358 \times 800$6–$1358 \times 800$7 for typical holograms. The mirror pitch of $1358 \times 800$8 sets the highest spatial frequency and induces an aliasing limit (Rocha et al., 2024, Bass et al., 7 Sep 2025).

Calibration is performed by choosing an illumination wavelength, optimizing the mirror-bias voltage to stretch the full phase span to $1358 \times 800$9, and then recording interferograms while the entire chip is set sequentially to each of the 16 codes. Two calibration geometries are described: an image-plane Twyman–Green arrangement with slight reference-arm tilt to produce straight fringes, and a Fourier-plane Young’s double-pinhole arrangement in which half the chip passes through one pinhole and half through the other. The phase response $10.8\,\mu\mathrm{m}$0 is extracted by fitting a sinusoid to fringe motion; the codes are then reordered so that $10.8\,\mu\mathrm{m}$1 is monotonic and stored in a lookup table (LUT) (Rocha et al., 2024).

3. Mathematical formulation for wavefront correction and tracking

In the adaptive-optics formulation, the incoming aberrated wavefront $10.8\,\mu\mathrm{m}$2 is expanded in a finite Zernike basis $10.8\,\mu\mathrm{m}$3 over the clear aperture $10.8\,\mu\mathrm{m}$4:

$10.8\,\mu\mathrm{m}$5

The objective is to choose a mirror phase $10.8\,\mu\mathrm{m}$6 that minimizes residual error. A convenient quadratic cost is the integrated squared residual phase,

$10.8\,\mu\mathrm{m}$7

In vector-matrix form, with $10.8\,\mu\mathrm{m}$8 denoting the sensed Zernike amplitudes and $10.8\,\mu\mathrm{m}$9, the same objective is written as

$95\%$0

where $95\%$1 is a weighting matrix, often the identity, and $95\%$2 is a scalar gain (Bass et al., 7 Sep 2025).

This formulation places actuator discreteness and update-rate limits directly into the control problem. The actuator constraint is

$95\%$3

with $95\%$4 levels. The temporal bandwidth is limited by $95\%$5, corresponding to a sampling period $95\%$6. Within this framework, simultaneous adaptive optics and tracking are expressed by superposition in the same transfer-matrix chain:

$95\%$7

A plausible implication is that the PLM is not treated merely as a programmable hologram source, but as a unified control surface on which low-order tracking and higher-order aberration correction share the same basis and hardware path (Bass et al., 7 Sep 2025).

4. Control workflows and real-time algorithms

Three real-time algorithmic regimes are explicitly described. The first is closed-loop adaptive optics with a wavefront sensor (WFS). The loop measures the Zernike vector $95\%$8 from Shack–Hartmann spots, computes the PLM phase vector according to

$95\%$9

uploads $95\%$0 to the PLM, and repeats at $95\%$1. For sufficiently small $95\%$2, specifically when $95\%$3, the loop is reported to be strictly stable and to reduce RMS error by approximately $95\%$4 (Bass et al., 7 Sep 2025).

The second regime is tip–tilt-only control using a camera or quad detector. Here $95\%$5 is replaced by a two-element vector, either $95\%$6 or $95\%$7. The transfer matrix $95\%$8 is remeasured for the new sensor geometry, and the same update law is applied with $95\%$9 modes. The third regime is smart search and acquisition under single-pixel feedback: the PLM is first defocused to enlarge the beam and execute a coarse angular scan, the highest-power direction is identified, the beam is refocused, and a fine spiral or conical scan is performed around the coarse direction; static higher aberrations may optionally be corrected by adding low-order Zernikes (Bass et al., 7 Sep 2025).

The streaming and software stack are described in detail for the DLP6750 EVM. A continuous-valued target phase profile $1358 \times 800$0 is quantized to the nearest LUT phase value, each 4-bit code is encoded in a $1358 \times 800$1 memory cell of binary pixels, and 24 such binary holograms are packed into the 8 bit-per-color channels of one RGB frame. The frame is sent over HDMI at 30 Hz or DisplayPort at 60 Hz to the PLM, which is recognized as a secondary monitor. On each video frame, the PLM unpacks the 24 holograms and displays them sequentially, each for $1358 \times 800$2 with a $1358 \times 800$3 inter-frame gap consisting of $1358 \times 800$4 return-to-flat and $1358 \times 800$5 step time, yielding an effective update rate of $1358 \times 800$6. External trigger lines provide one pulse per hologram for synchronization with cameras or other hardware. The accompanying C++ library, “plmctrl,” wraps low-level graphics APIs such as DirectX 11 and provides functions including initPLM(), setMirrorBias(), uploadRGBFrame(), startStreaming(), stopStreaming(), and getTriggerLine(); bindings were tested under Windows in Python, MATLAB, and LabVIEW (Rocha et al., 2024).

5. Reported performance and demonstrated use cases

The adaptive-optics and tracking study reports that closed-loop correction of the first five Zernikes reduced RMS wavefront error from several wavelengths to less than $1358 \times 800$7 in less than $1358 \times 800$8 convergence time. In tip–tilt tracking, the residual pointing error was less than $1358 \times 800$9 for $10.8\,\mu\mathrm{m}$0 light at update rates up to $10.8\,\mu\mathrm{m}$1. For beam coupling into a single-mode fiber, automated reacquisition in less than $10.8\,\mu\mathrm{m}$2 was obtained using conical scan plus static aberration correction, achieving greater than $10.8\,\mu\mathrm{m}$3 coupling efficiency (Bass et al., 7 Sep 2025).

The wavefront-shaping study reports several optical demonstrations. In arbitrary pattern projection, a holographic Gerchberg–Saxton design projected in the Fourier plane at $10.8\,\mu\mathrm{m}$4 with $10.8\,\mu\mathrm{m}$5 and demagnification $10.8\,\mu\mathrm{m}$6 steered approximately $10.8\,\mu\mathrm{m}$7 of incident power into the synthesized image, with contrast greater than $10.8\,\mu\mathrm{m}$8. Measured first-order diffraction efficiency remained above $10.8\,\mu\mathrm{m}$9 up to diffraction angles $\approx 95\%$0 when using a linear-ramp hologram with approximately six pixels between $\approx 95\%$1 phase wraps (Rocha et al., 2024).

For aberration measurement and compensation, the PLM was divided into $\approx 95\%$2 super-pixels of $\approx 95\%$3 mirrors each. The phase of each super-pixel was measured interferometrically, requiring $\approx 95\%$4 holograms, or $\approx 95\%$5 at $\approx 95\%$6. Conjugating the measured aberration map and adding it to a tilt hologram yielded a diffraction-limited focus with $\approx 95\%$7 higher peak intensity. Native chip-curvature aberration was reported to span several interference fringes, approximately $\approx 95\%$8 rad over the full aperture, while in-situ correction reduced residual wavefront error to below $\approx 95\%$9 RMS (Rocha et al., 2024).

High-speed wavefront shaping through a multimode fiber was demonstrated with the sequence PLM $1.09$00 10$1.09$01 objective $1.09$02 1 m step-index MMF with $1.09$03 and core radius $1.09$04, with the fiber output imaged on a high-speed camera. Transmission-matrix measurement employed 2680 super-pixels of $1.09$05 mirrors and four-step phase-shifting holography, requiring 10,720 holograms or $1.09$06 at $1.09$07. The computed input field was $1.09$08, with amplitude and phase encoded using the PLM’s 4-bit profile. The achieved focus had a power ratio $1.09$09, and 2160 focus positions were scanned at $1.09$10 by streaming 90 full 24-hologram frames. The abstract summarizes this regime as scanning over 2000 points at $1.09$11 (Rocha et al., 2024).

6. Relation to other modulators, limitations, and system implications

The PLM is positioned between several established spatial-light-modulation technologies. Relative to liquid-crystal SLMs, the reported comparison emphasizes that LC devices offer high efficiency, bit depth from 8 to 16 bit, and megapixel resolution down to $1.09$12 pitch, but are typically slow at 60–120 Hz, with faster models near $1.09$13 via heating or overdrive, and ferroelectric devices above $1.09$14 but with binary phase and less than $1.09$15 efficiency. LC-SLMs are also polarization-sensitive, and pixel crosstalk and flicker can degrade fidelity. Relative to deformable mirrors, the comparison emphasizes multi-wavelength stroke and speeds up to a few kilohertz, but only a few hundred to a few thousand actuators and limited suitability for high-resolution holography. Relative to DMDs, the comparison emphasizes switching above $1.09$16 but binary amplitude-only control, less than $1.09$17 useful diffracted power for phase tasks, many spurious orders, and reduced degrees of freedom (Rocha et al., 2024).

Within that comparison, the PLM is described as combining high efficiency above $1.09$18, polarization-agnostic phase-only control, 1 MP resolution, fast sub-kHz full-frame updates, and direct 4-bit phase on every pixel; it is also described as compact, as leveraging existing DLP infrastructure, and as scaling to planned $1.09$19 operation. At the same time, the reported trade-offs are explicit: full WFS correction with more than five modes provides the best Strehl but requires larger SWaP and more compute; camera- or quad-based tip–tilt-only control with $1.09$20 minimizes SWaP and supports very high update rate but only modest alignment accuracy of about $1.09$21; and smart scanning adds a few extra microseconds per step while eliminating all external beam-steering optics (Rocha et al., 2024, Bass et al., 7 Sep 2025).

A common misunderstanding is to equate the mirror step response with currently available continuous full-frame streaming. The device-level characterization reports single-step mirror response below $1.09$22 and a fundamental modulation bandwidth of approximately $1.09$23, but current electronics limit continuous streaming to $1.09$24; firmware-enabled bursts up to $1.09$25 have been demonstrated, and operation up to $1.09$26 is anticipated. Another misconception is to treat the PLM as a purely beam-steering component. The adaptive-optics study instead formulates it as a monolithic phase modulator whose single pattern can superpose tracking and higher-order AO commands, with time-multiplexing between high-order AO updates and rapid tip–tilt updates to share the 1–5 kHz bandwidth (Rocha et al., 2024, Bass et al., 7 Sep 2025).

Taken together, the reported results suggest a distinct role for the Texas Instruments PLM: a monolithic, chip-scale phase modulator that can unify high-resolution wavefront shaping, adaptive-optics correction, beam tracking, and acquisition workflows within a common Zernike-to-pixel transfer framework, while replacing deformable mirrors, steering mirrors, and separate acquisition subsystems in low-SWaP optical systems (Bass et al., 7 Sep 2025).