Optimized Reticle Dispatch Improves Throughput

IBM's Building 323 in East Fishkill, N.Y., which Semiconductor International last year awarded with the Top Fab designation,¹ is a fully automated 300 mm fab with an impressive array of automation capabilities. With 15,000 dispatches a day, all production, development and engineering lots are dispatched to tools without human intervention.

The fab's photolithography environment includes more than 24 integrated photo tools, with tools located across multiple bays, and distributed reticle storage (one pod stocker and one or two bare reticle stockers per bay). Achieving full automation also requires reticle automation, and as such the 3000 reticle moves a day at the East Fishkill fab are also based on decisions made by the automated dispatching system. Failure to manage and deliver reticles in an accurate and timely manner would negate many of the advantages associated with automated material handling system (AMHS) FOUP automation.

In a presentation at the recent ISMI Symposium on Manufacturing Effectiveness in Austin, Texas, Rich Burda, 300 mm industrial engineering at IBM Microelectronics, detailed the chipmaker's efforts to optimize its reticle coordination within its automated 300 mm fab, identifying best practices in its photolithography operations and its reticle operations. Key performance criteria include making sure the reticle is at a given tool prior to lot dispatch, lot sequencing considerations based on the reticle and track recipe, a reticle qualification strategy, and minimized reticle movement. “We want to dispatch how we want and not worry about the reticles,” Burda said.

An automated reticle handling system (ARHS) vehicle is poised at a reticle pod stocker. (Source: IBM Microelectronics)

Although the initial automated reticle handling system (ARHS) release was successful, there was room for improvement, according to Burda. The engineers still saw some idle with work in progress (WIP), especially with low-WIP situations, he said. The rules were not designed to make tool-to-tool moves from a busy tool to an idle tool. So the next step focused on a new rule that would identify imbalances to move jobs from a busy tool to less busy tool. The reticle rebalance includes a logic progression that first generates a simplified lot dispatch list for the tool group and compares that with tool reticle inventories; identifies tools in need of reticles (idle or at risk of becoming idle); identifies tools with a surplus of reticles; and identifies candidate reticles for tools in need and tools in surplus.

Within a 24-hour period, the ARHS executes ~100 of these tool-to-tool moves, reducing reticle-related “idle with WIP” by ~20%. However, there were still some logistical deficiencies: Between-lot delays were limiting PPH; reticle movement was always a reaction to current WIP; lot/reticle dispatching heuristics required adjustment to match changing operator conditions; and the rebalancing rule did not completely fix balancing issues.

The reticle scheduling solution was to use short-term, local-area optimization to schedule the lot/reticle flow. This solution — implemented in production on all lithography tools last quarter — coordinates the lot schedule with the reticle schedule. “Instead of looking at one tool, it's making a decision for the entire tool group,” he said. “In rebalancing, it would have to make the wrong move first and correct. Here, it didn't have that need.”

The objective is to fill load ports, but there are times when reticle load ports will not be filled, Burda noted, giving the chipmaker increased flexibility. The key result is improved throughput with optimized reticle/recipe sequencing. For the first time, IBM is able to deliver reticles to a tool before it's in the queue.