Principle of optical spherical symmetry
As shown in Fig. 1a, the first method of spherical symmetry technology uses two concentric spherical lenses, L1 and L2, with the human eyeball positioned near the center. To maintain spherical symmetry, a transparent spherical screen concentric with L1 and L2 is required. Currently, flexible screens struggle to achieve transparency at high resolutions26 and to conform to spherical surfaces. Directly manufacturing high-resolution spherical screens involves innovations in semiconductor processes, such as photolithography, which are extremely challenging to industrialize. To address this, we propose using a rotating arc screen to achieve a transparent spherical screen. Rotating screens, also known as persistence of vision (POV) display27, are typically used for direct advertising displays. We miniaturized this technology and applied it to AR glasses for the first time. The advantage of rotating displays is their inherent transparency. Additionally, as line displays, the pixel count is reduced by three orders of magnitude compared to matrix screens, eliminating the issue of massive transfers28-30 and reducing the difficulty of achieving high resolution.
In Fig. 1a, the top left corner shows the polarized light path, aiming to shorten the overall system thickness. Arc-shaped displays (AD) are typically micro-LED arrays. The emitted unpolarized light first passes through a linear polarizer (LP) and a quarter-wave plate (QWP), becoming left-handed circularly polarized (LCP) light. After passing through L1 and L2, it goes through a QWP on the left side of L1, becoming P-polarized light, which is reflected by a reflective polarizer (RP) film. After passing through another QWP, it becomes LCP light again. The outer surface of L2 is coated with a semi-reflective film. Light hitting the outer surface of L2 changes from left-handed to right-handed circularly polarized (RCP) light, and after passing through L1 and the QWP again, it becomes S-polarized light, which then passes through the RP film into the human eye. The nearest spherical radius to the eye is 21 mm, the farthest is 32.3 mm, and the AD radius is 33 mm, with a total optical thickness of 13mm. It is clear that the optical paths for the central and edge FOVs are the same, so if the center is clear, the edges will naturally be clear as well, and the FOV is no longer limited. Additionally, the system only has spherical aberration and minor chromatic aberration, without coma, field curvature, astigmatism, or distortion, thus achieving very high optical resolution. As shown in Fig. 1b, we calculated the MTF value of the optical system at an Eyebox of 8 mm. Based on the arc length, we determined the pixel sizes corresponding to image resolutions of 2K, 4K, 6K, and 8K, as well as their respective Nyquist frequencies. For a 90 ° FOV, based on the human eye's resolution limit of 1 arcminute, the maximum required resolution is 90 × 60 = 5400, approximately 6K. Our optical system's MTF value is greater than 0.5 at this resolution across all FOV values, indicating that the system can display sharp and clear images even at the human eye's resolution limit. Beyond the 8 mm region, the resolution gradually decreases. However, since the pupil diameter of the human eye can average 3mm under normal brightness, when we scan the edge of a 10 mm Eyebox with a 3 mm pupil. We find that the MTF value remains greater than 0.2@40lp/mm, supporting 4k clarity at the edges. Within the 12 mm Eyebox range, the MTF value remains greater than 0.2@20lp/mm, supporting 2k clarity at the edges.
To further reduce the overall thickness, since the human eye's pupil is generally larger than 2 mm, when the width of the rotating display is smaller than the pupil, if placed between L1 and L2, after multiple reflections, light will still enter the eye, as shown in Fig. 1c. Clearly, the smaller the width of the rotating display, the better. We fabricated strips of different widths and tested them at a rotation speed of 3600 RPM, finding that when the width is less than 0.5 mm, the human eye cannot detect its presence. The upper left corner of Fig. 1c shows the polarization optical path diagram. The light emitted by the AD first passes through a LP and a QWP, becoming RCP light, which is then reflected by L2 to become LCP light. This light then passes through L1 and the QWP to become P-polarized light, which is reflected by the RP film. The reflected light passes through the QWP again, becoming LCP light, and is once again reflected by L2 to become RCP light. This reflected light passes through L1 and the QWP again to become S-polarized light, ultimately passing through the RP film and entering the human eye. Compared to Fig. 1a, the light in this system can be folded one more time, reaching the theoretical limit of foldable times in RP system. The overall thickness is reduced to 8.6 mm. Additionally, L1 and L2 can serve as protective lenses to safeguard the rotation of the AD, eliminating the need for additional protective structures. As shown in Fig. 1d, the wearing effect of this structure is similar to ordinary sunglasses, being very lightweight and thin.
In the above structures, since the lenses are concentric and almost afocal, ambient light can pass through the system to the human eye with minimal impact, achieving an OST function. The curvature of the surface of L1 near the eye can be specially designed according to the user's myopia or hyperopia degree, thus correcting both ambient light and virtual images simultaneously.
Electronic system and mechanical structure
To validate this unprecedented optical design, we undertook a comprehensive series of fabrication and manufacturing processes to develop a functional prototype. This involved integrating advanced electronic systems and precision mechanical structures to demonstrate the feasibility and performance of our spherical symmetric AR glasses.
Compared to panel displays, rotating displays have faster sub-pixel refresh rates. For example, to display 60 frames per second with a rotational resolution of 2048, the response time for a single pixel is 1/(60×2048) = 8.14 μs. This eliminates the possibility of using LCDs as display pixels. OLEDs, with their fastest response time around 10 microseconds31-32, cannot meet the higher resolution requirements and also have lower brightness. Therefore, the only candidate is micro-LED, which has a response time of 0.2 nanoseconds32 and can meet extremely high-resolution requirements. For instance, at 120 frames per second and a rotational resolution of 5400, the required response time is 1.54 microseconds, well within the LED's limit. Even with 255 grayscale PWM dimming, the required response time is 6 nanoseconds, still within the micro-LED’s response range. Additionally, LEDs can provide brightness levels exceeding one million nits33, which is crucial for rotating displays.
To achieve optimal performance, CMOS technology34 must be used to integrate both LED digital drivers and analog drivers onto strip-shaped silicon-based chips. To create a transparent spherical screen, micro-LEDs must be arranged along an arc. Currently, there are at least two methods to achieve this:
1. Ultra-Thin Silicon Wafer Method: The drive circuitry is fabricated on flexible, ultra-thin silicon wafers35,36. These wafers are then cut into straight strips, which can be bent into the desired arc.
2. Micro-Reflector Method: As shown in Fig. 2a, the drive circuitry is directly fabricated into an arc on a conventionally thick wafer. Light is then bent 90 degrees using micro-reflectors to emit from the side.
After completing the CMOS substrates using either method, micro-LEDs can be positioned correctly through a transfer process. Only 3×2000=6,000 pixels are needed, refreshed through rotation, to achieve a 4K resolution RGB full-color screen. This considerably reduces the complexity of mass transfer28 compared to traditional panels, which require about 27 million pixels. Consequently, the number of strips that can be cut from the same wafer size is much greater than that of rectangular panels, significantly lowering the screen cost.
Currently, we focus on the successful prototype manufacturing, then move to the next phase with high development costs and long cycles of silicon-based chips. For prototype manufacturing, we use two LED control chips, HD107S, wire-bonded to a PCB substrate, allowing subsequent SMT soldering onto the light strip. This chip features 8-bit grayscale, a channel refresh rate of 37 KHz, a maximum current of 17 mA per channel, and a data clock of 25 MHz. As shown in Fig. 2b, we use a 0.17 mm thick BT board to create a cross-shaped light strip (CLS), with mini-LEDs size of 100 μm by 200 μm soldered onto it using a die-bonding process. The mini-LED control chip is on the back of the CLS. The control logic, as shown in Fig. 2c, allows each chip to independently control 6 LEDs, interconnected via SPI control and clock signals to control all mini-LEDs on a single arm. The pixel pitch between mini-LEDs is 240 μm, with each arm containing 96 LEDs, totaling 384 LEDs. As shown in Fig. 2d, the positions of the LEDs on the four arms are specially designed to complement each other’s rotational radii, ultimately reducing the image pixel pitch to 60 μm. Since the LED size is larger than 60 μm, this arrangement causes overlap between LEDs, reducing image resolution. To eliminate this overlapping effect, we manufacture precise masks and then coat them with chromium to cover areas beyond the 60-micron diameter range. This allows light to pass only through the 60-micron holes, thereby further reducing the size of the LED lights. Finally, the radial resolution achieved through this complementary arrangement is 768. In the rotational direction, the resolution achieved is 2048, resulting in an overall image resolution of approximately 768×2048 for a single green channel. Due to LED control chip limitations, the maximum frame rate at this resolution is 20 FPS. As shown in Fig. 2e, the mainboard is ring-shaped to allow light to pass through the center. It integrates an FPGA, SD card, wireless MCU, encoder chip, and four FPC connectors. The FPC connectors link the CLS to the mainboard. The FPGA sends data stored in the SD card to the LED driver IC according to the required rules to drive the LED refresh display. The wireless MCU is used for remote system debugging. The encoder chip accurately obtains rotational angle information and provides it to the FPGA to control the Micro-LEDs' image refresh at that angle. As shown in Fig. 2f, the high-precision film code disk we use has 1024 equally spaced reflective strips, with one longer strip indicating the initiation point, or the image's starting position. The encoder chip supports up to 4x frequency multiplication, achieving an angular resolution of 4096 with this code disk.
Due to the size limitations of the existing LED control chips, the light strips used in this article have a width of 4 mm, exceeding the size of a normal human eye's pupil. Therefore, we can only adopt the optical structure shown in Fig. 1a, placing the light strips outside L1 and L2. To achieve a precise spherical surface, we used CNC to create a precision bracket with a spherical inner surface. The CLS is closely attached to the surface of this bracket, bending it into a spherical shape, which presents a spherical screen effect when rotating. As shown in Fig. 2g, the light strips are embedded into the bracket to form a sphere when rotated. The structure of a single eye is shown in Fig. 2h, with an additional protective lens on the outermost side to prevent the rotating object from contacting the outside. On the side close to the human eye, we use a PCB coil to drive the permanent magnet attached to the mainboard. The PCB itself serves as part of the glasses housing and also functions to drive the rotation of the light strip. At the bottom of the PCB, we attach a ferrite film to improve motor efficiency and generate attraction to the rotor, limiting its movement in other directions.
Since the solution in this article employs a rotating structure, achieving low-friction, high-precision rotation is crucial. Traditional ball bearing structures produce significant noise, and a single bearing generally weighs more than 5g, making it unsuitable for this system. Air bearings37 are frictionless and noiseless, widely used in gyroscopes. However, they generally require speeds of tens of thousands of RPM and have poor rigidity, making it difficult to apply them to this system. Magnetic levitation bearings38 might be a solution, but their control systems are complex, and the rotor runout is significant, making them not an ideal solution currently. Since our system can be a sealed space, it can be filled with inert gases such as nitrogen, providing the possibility of using superlubricity39,40 materials. We deposited a DLC film41 on precision-machined stainless steel using the CVD method, with a deposition environment of 75% hydrogen and 25% methane. Under these conditions, the friction coefficient is reduced to 0.001. This allows our rotational power consumption to be <100mW@1500RPM, with noise less than 30 dB. Through ultra-high precision machining, the gap between the rotor and stator has been controlled within a range of 2 μm, ensuring very small runout, thereby achieving high-precision display.
To summarize, Table 1 shows the key parameters the system can support.
Table 1. The key optical parameters and characteristics of the system
Parameter
|
Value/Description
|
FOV
|
>90°
|
Eyebox
|
>10 mm
|
Resolution
|
Supporting 8K
|
MTF
|
> 0.5 at 6K resolution across all FOV
|
Screen Type
|
Rotating arc screen (POV display)
|
Minimum Optical Thickness
|
8.6 mm
|
Aberration characteristics
|
Spherical aberration, minor chromatic aberration. No coma, field curvature, astigmatism, or distortion.
|
OST Function
|
Minimal impact on ambient light passing through the system to the human eye.
|
Vision correction
|
Curvature of L1 surface near the eye can be specially designed to correct myopia or hyperopia, correcting both ambient light and virtual images.
|