Preliminry feasiblity analysis of using JBD AMuLED display as maskless lithography image source
1. Assumptions
Stepper: ASML PAS 5500 (This is the closest to PAS5000 from HKUST, for which data is available)
Exposure intensity: 900mW/cm2
Reduction ratio: 5:1
Exposure time: 250ms (estimated, this yields 225mJ/cm2, more-or-less in range for HPR504)
Resist: HPR 504 (positive DQN-based)
Dose-to-clear: 250mJ/cm2 (for g-line. Based on literature, sensitivity at 400nm is not lower than this)
MFS: 1um, resolution (lambda): 500nm
Maximum targeted die size: 5x5mm
It is assumed that we have one pixel per lambda.
It is also assumed that approx. 5% of optical output from the display is focused onto the wafer, the rest is either falling outside of the optic's acceptance angle or is wasted by the optics.
2. JBD AMuLED display
Very little information is available on the display itself. From information publicly available, it can be deduced that the JBD Active-matrix micro-LED display is a silicon-monolithic IC consisting a silicon backplane on which LED drivers with per-pixel PWM modulation are present, and a light emitting layer based on III-N compound semiconductor LEDs. According to [1], this is realized by growing GaN or other III-N epitaxial layers on sapphire wafers like in common LED production, then bonding this wafer to the pre-processed Si wafer, then separating the sapphire substrate, leaving the epitaxial layer behind for further processing.
According to JBD website, the displays have high uniformity in emission by means of an on-backplane NVM for storing intensity compensation data.
For the following statements, the following devices were considered:
JBD5UM720P-UVA
JBD25UMFHDB
Based on the metrics of these devices, the following can be inferred for a hypothetical custom display IC:
Optical power: 70nW/pixel
Pixel pitch: 2.5um
Wavelength: 405um
WPE: 2.7%
3. Wavelength, coherence, intensity and uniformity
Based on the information above, it can be inferred that the LED layer stack of JBD5UM720P-UVA is capable of emitting at 405nm, to which HPR504 is sensitive. Since the source is composed of individual pixels instead of a single source with a photomask, the illumination will be inherently incoherent. Its effects on projection is to be assessed. Uniformity can be ensured, as the display accepts grayscale image.
The intensity at the display is 70nW/pixel for 5.25um2 pixels, that yields 1300mW/cm2. After 5:1 reduction and consideration of 5% will pass trough the optics, this is 1625mW/cm2, sufficient to expose HPR504 in 150ms.
4. Resolution, display size, defect density
It is assumed that 5kDPI and 10kDPI displays use similar LED layer stacks, so if a wavelength is achieveable on one pitch, then it is achieveable on the other as well. Based on that, let's take a hypothetical display emitting at 405nm with a pixel pitch of 2.5um. After 5:1 reduction, this yields a resolution of 500nm, the same that was used for Pearlriver masks in HKUST, therefore a 2.5um pitch is sufficient for a 1um MFS technology.
For a die size of 5x5mm, which is at 1um sufficient for low-performance embedded devices (example: PIC18F4550, 40MHz 8-bit CPU wo. MMU, 4K SRAM, 64K flash, 10-bit ADC, USB on a 4x4mm 0.8um die), the display size (active area) would be 2.5cm x 2.5cm, 100Mpx. It may be challenging and expensive to manufacture such large chip with acceptable defect density.
Since the pixel pitch is the half of the MFS and both brightfield and darkfield exposures are expected, no dead pixels are acceptable on the display, neither stuck-on nor stuck-off (therefore, no deadpixel-compensation like redundant cutoff can be applied). This would require a defect density lower than 10ppb. JBD currently states its defect density is in the ppm range.
5. Integration proposal
Since adoption of LS1U and further LS nodes at commercially viable production scale (i.e. outside of R&D labs) is paramount for reaching the goal of the Libresilicon project (providing really affordable near-single-quantity custom IC production for SMEs, startups, and last but not least, individuals), and the strongest blocking point in such adoption is the cost of obtaining and operating maskless photolithography equipment (all other steps require standard equipment, and easily transferable recipes are published by the project), I recommend to implement the image source unit in LS's maskless stepper "in the form of a glass mask", that is, in such way that its mechanical and optical interface is compatible with existing steppers. For example, the imaging unit may rest on the surface of a 5-inch rectangular glass substrate, with its sides chrome-plated and patterned with alignment marks compatible with ASML steppers, with a UV sensor at the backside is used as a gating input for the display, allowing the unit to be dropped into an ASML (or whatever other brand) stepper without requiring time-consuming, money-burning, warranty-voiding modifications to the adopting fab's existing stepper itself.
Technical problems to be solved with this arrangement are:
a) The imaging unit is expected to generate considerable amount of heat (6.25cm2 with 1300mW/cm2 intensity output and 2.7% WPE is approx. 300W power consumption, most of which is dissipated), resulting in considerable thermal expansion, that needs to be managed (cooling, or using low-LCTE materials).
b) Power supply and data connections need to be implemented without jeopardizing mechanical compatibility.
6. Conclusion and open points
The use of JBD's AMuLED display as maskless lithography pattern source is not infeasible outright.
Points to be further evaluated:
- Effect of incoherent illumination on projection optics
- Evaluation on JBD side if they want to develop a 2.5x2.5cm unit for us...
- ... and How Much Does It Cost?
- Defect density - mitigation or enhancement
- Feasiblity analysis of the proposed integration on INL/LS side
