Homemade semiconductor · lab notebook, in progress

Maskless Lithography Bench

A consumer 4K resin printer's UV screen makes a photomask that can be redrawn at will, so there is no physical mask. Light through the lit pixels exposes photoresist on a 4" silicon wafer, an ultrasonic bath develops it, and a microscope checks what is left. Below is the bench with the near walls of the printer and the bath cut away. The calibration lines print; devices do not, yet.

MASKLESS · 35 μm
UV through the lit pixels The wafer, wherever it is in the cycle Developer bath The near walls of the printer and the bath are cut away; cut faces are white. On the printer the LCD's pixels are drawn as cells about 90 times their true size so the bars can be counted. The round view is the microscope's, to scale: its field is about 1 mm across.

What the lab is tuning

10Calibration line3 px · 105 μm
The screen can only light whole pixels, about 35 μm each, so a 100 μm line is drawn 3 pixels wide (105 μm) and a 250 μm line 7 pixels wide (245 μm).
Relative to the working setting. Too little and the lines do not clear; too much and they bloom wider than the pixels that drew them.
Relative, like the dose. Cut short, resist is left in the lines.
07Ultrasonic agitation Keeps fresh developer moving over the wafer. Without it the lines clear unevenly.

Parts of the bench

Taken from the lab notebook

  • Maskless stereolithography. A consumer 4K mSLA resin printer's UV LCD is the programmable photomask. Its pixel is about 35 × 35 μm.
  • The wafer. 4" P-type, boron-doped silicon, 500 ± 25 μm thick, with a 300 nm thermally grown SiO₂ layer. It arrived in a clear eight-arm wafer carrier.
  • The process. Photoresist on the wafer is exposed through the LCD, then developed in an ultrasonic sodium carbonate bath for even agitation. Exposure time, development time and agitation are what is being tuned.
  • The evidence. A calibration pattern of 100 μm and 250 μm lines, photographed under a microscope, resolving at about 100–105 μm.
  • The status. Calibration is printing; devices are not. The features are two orders of magnitude above the 50–1000 nm thesis. This lab builds process intuition; it does not claim a node.

Added for this figure

  • No recipe, on purpose. Dose and develop time are relative to a working setting (1.0×), with no seconds. The figure gives no concentrations, temperatures or how-to steps.
  • Pixel snapping is arithmetic: a line is the nearest whole number of 35 μm pixels, so 100 μm becomes 3 px (105 μm) and 250 μm becomes 7 px (245 μm).
  • Schematic, not measured: how underdosing leaves lines not cleared or ragged, how overdosing blooms them wider, and how lines clear unevenly without agitation. The direction is right; the amounts are not predictions.
  • The screen is drawn as 3840 × 2400 pixels at 35 μm, 134 × 84 mm. The notebook doesn't name the panel. Its cells are drawn about 90 times too large, and the wafer is drawn twice its true thickness.
  • The bench's own parts: the carrier's chuck and edge clamps, the lens plate, the dipper and its stand, the bath's controls and the microscope's form are generic. Wafer handling between stations is shown as a cross-fade.
  • The cycle's pacing is the animation's own and stands for no real times.

Source: Atharva Rajé's lab notebook, in progress, and the microscope photograph of the 100 μm / 250 μm calibration pattern.