Virtual fabs give industry a higher reach Virtual fabs give industry a higher reach

11/01/1997

Virtual fabs give industry a higher reach

The semiconductor industry has picked the low-hanging apples of wafer yield. Now, with the advent of 300 mm, achieving success is a farther reach. The industry may experience some of its most challenging times yet due to the gargantuan task of 300-mm production, coupled with an added number of wafers in process, and the increasingly stringent demands for environmentally sustainable manufacturing. On the plus side, the 300-mm transition, because of the retooling involved, will let many companies "do the planning right" this time for their new fabs.

To succeed, fabs will need to hone in on productivity gains anywhere they can be found. Fabs will have to rely on more automation - more computer hardware, software, simulations, and modelling. Heuristic approaches to semiconductor manufacturing just won`t produce good results anymore.

A macro approach to the fab line will also be required - individual optimization of tools will no longer succeed. It`s a lesson learned from the heyday of operations research when students were shown that in the classic constrained optimization problem, maximizing each variable does not necessarily maximize the objective function.

Semiconductor manufacturing, viewed as a classic constrained nonlinear optimization problem, is intractable. To optimize production by trying a closed solution and wringing out the last percentage points based on the individual variables is out of reach. Even though we have the computing power to do this task, we still do not have a viable theory. Computationally accessible does not mean computationally practical.

The industry rises above this problem by using modelling and simulation. Certain software programs can simulate the entire production line with all the fixed and probabilistic parameters needed to show its operation. The simulation, however, is not a solution to the productivity problem. It provides an anecdotal optimization. A scenario is run for a production shift based on some wafer/hr assumption and a certain product mix, which ultimately reveals the bottlenecks. Then engineers can tweak variables and play "what-if" games to try to eliminate these bottlenecks. But they have no idea whether their efforts come close to a global optimal solution.

However, simulation and modelling are our best tools yet because they make the virtual fab concept possible. This concept, which has been around for some time, is used by several companies in their product descriptions. Some models of this concept just treat parts of the total process, such as TSMC`s total order management system, which schedules tools to better meet customer delivery dates.

Maintaining and developing this virtual-fab mentality will be essential in the new generation of fabs. For example, in Amkor/Anam`s soup-to-nuts Seoul fabrication plant, this type of thinking is at the forefront of the company`s plans - the idea lends itself readily to foundry operations.

With the virtual fab concept, the customer can participate as if he owns a piece of the fab. Amkor/Anam will have a control room where the customer can view and monitor his particular process. Thus, several customers can view the fab as their own, even as others are simultaneously doing the same thing. This process is analogous to computer time sharing or multiplexing, where each client believes he is the only one using the system, yet thousands of others are allowed access in their own slice of time.

Until we somehow develop more deterministic ways of dealing with the tens of thousands of variables present in a modern semiconductor fabrication production line, the virtual fab will allow us to pick and savor the higher apples on the tree.