Hybrid Cluster Tool Architectures for 300mm: A Flexible, Cost-Effective Solution

By Dan Camporese, chief scientist, Brooks Automation Software Corp.

Hybrid cluster tool architectures which combine atmospheric front-end buffer stations with vacuum-based cluster tools are now being proposed for 300mm fabs. A hybrid cluster tool provides WIP (work in progress) buffering in front of the cluster tool, increased tool utilization, and automated factory interfacing between the cluster tool and FOUP (front opening universal pod) delivery systems. The processing of multiple cassettes from the buffer station, through the tool, and back again in this type of hybrid architecture poses many controls and automation challenges. Can the hand-off between buffer station and cluster tool be managed by the cluster tool's own CTC, or should it be handled via a single TMC (transport module control) model?

Semiconductor processing equipment comes in all shapes and sizes, but developers of controls software classify tools into one of two basic types: batch or cluster tool. A traditional batch tool accepts a cassette of wafers and processes all wafers in a single processing module at one time (e.g. diffusion furnaces, epitaxial reactors). With the transition to larger 200mm wafer sizes, single wafer processing tools have become common. In these systems, multiple process modules, each independently processing individual wafers, cluster around a single wafer handler.

Batch tools use centralized control systems. Monolithic control architectures, used in early cluster tools, have proved unwieldy as process module complexity has increased and customers have demanded more flexible cluster configurations. Present day cluster tools typically have distributed control architectures with independent control computers for the transport subsystem and each individual process module, all coordinated by a centralized cluster tool controller. Brooks supplies the "middleware", or the hardware and software portions of a cluster tool between the factory interface and the physical process modules.

300mm fab factory interface requirements

From a fab's perspective, the cluster tool is a place where cassettes of wafers are delivered for processing and removed after processing. The factory-to-tool cassette exchange contributes to the tool's effectiveness. For 300mm fabs, the trend is away from manually loaded open cassettes to enclosed FOUPs, automatically delivered via automated guided vehicles (AGVs) or overhead track robots, as specified by 300 mm standards committees—I300I and Selete.

As the automated delivery system will service a number of tools in a bay, it must be able to perform exchanges with the tool at its convenience, rather than at the tool's demand. This requirement creates a need for buffering at the tool. If the tool has two cassette stations, one station can be available for exchange with the factory while the other cassette is being processed.

To maximize tool productivity and avoid having a tool wait for an upstream cassette to be processed, one can use an expanded atmospheric buffer station to hold and manage four cassettes. Increasing the number of cassette station sites at the tool increases WIP capacity, reduces the need for other factory storage sites, and maximizes productivity of high throughput tools. As an alternative, one or two of the four cassette stations can be used for test, qualification, or clean wafers. By separating these wafers out of the production cassette, their frequency of use may be set dynamically based on tool requirements, rather than on a global factory-wide schedule.

The hybrid cluster tool option

The transport subsystem of a typical cluster tool includes a mechanical factory interface for cassette exchanges with the factory, and a robotic wafer handler for delivery to and removal from individual process modules. The load lock is the traditional interface between the factory environment (atmosphere), the tool environment (vacuum), and the cassette exchange point. The load lock is large enough to accommodate the entire cassette, which is then pumped to vacuum to avoid contamination between the factory and tool environments. The robotic wafer handler selects wafers, one at a time, from the load lock for processing and returns them to the appropriate slot.

A proposed hybrid architecture for 300mm cluster tools combines a highly controlled particle-free vacuum processing environment for wafers, with a flexible interface for cassette exchanges with the factory. This architecture integrates a vacuum transport system with a two- or four-cassette front-end atmospheric buffer station. This configuration decouples the factory interface from the process environment and reduces the required load lock capacity from one cassette to one or more wafers. Two robots, one at atmosphere and one at vacuum, operate semi-independently, but are coordinated through a sophisticated cluster tool control computer.

This architecture increases buffering flexibility at the tool, provides automated interfacing for FOUP delivery systems, maximizes tool utilization, and is more cost effective. Adding additional atmospheric buffer cassette stations at the front end of the tool costs considerably less than increasing the number of load locks on a vacuum cluster tool.

Modeling the control software

While the hybrid cluster tool architecture offers several advantages, it creates new challenges for the cluster tool control system. Ideally, the CTC should view the atmospheric and vacuum transport components as separate entities with explicit routings of wafers through internal load locks. This design is still controversial, and alternative, simpler models for the transport system are being contemplated. By exploring some of these simpler options, we hope to show the validity of the initial statement.

Single transport model

A hybrid cluster might be modeled as a single robotic system by hiding the actions of the atmospheric robot inside of transport operations. In one approach, the CTC treats the transport system (TMC) the same as a two-load-lock vacuum cluster tool, but finds that the Cassette Load and Unload operations are slower than a conventional cluster without the buffer front-end. The disadvantages of this approach are:

• The CTC will view the system as having only two load ports even though the actual buffer station may offer four factory interfacing ports.

• The unmodeled ports could be used to hold test and clean wafers, but any introduction of test wafers into the load lock would have to be done "magically" by the transport system without the CTC knowing.

• If the CTC only understands the existence of two load ports, it can only offer a two-load-port tool model to the factory over the GEM interface.

• Failure in the middle of a buffer-to-load-lock transfer would need to be recovered without any help from the CTC.

Another alternative is to hide the load locks from the CTC so that it directs wafers to move directly from the cassette station in the buffer module to a process module attached to the vacuum transport chamber. The coordination of the move through the load lock would be hidden from the CTC and handled entirely by the transport system. This approach would introduce a lot of transport "dead time" while the robots wait for the load lock pump and vent cycle operations which would occur inline with any move the CTC specified. To avoid this dead time, the transport system controller would have to be able to disassemble a series of end-to-end moves into component parts and reorder them to maximize parallelism. The use of a queue of end-to-end moves as the interface between the CTC and TMC is highly problematic.

Experience gained in this latter approach in early cluster projects led us to abandon it when designing Brooks' ClusterLink 3 product. Recovery from transport failure became virtually impossible as the CTC was unable to understand exactly where in the move queue the failure occurred after the TMC had reordered atomic components within each move. Flushing the queue and resending it was not viable as some moves may have been partially executed and needed an executive decision as to whether they should be completed or abandoned.

Dual transport model

The least problematic software model for the hybrid cluster tool, as seen by the CTC controller, appears to be one which includes the two robots and internal load lock resources explicitly. The load locks are modeled as process modules with two access ports, one accessible by each robot. The CTC scheduler/optimizer accommodates the load lock pump and vent times as process times provided to the CTC. A route (or flow) specification for a particular process would include the load locks as visits at the beginning and end of the route. The route could specify one load lock to always be used for entry and the other for exit or define the two load locks as parallel visits, thus leaving optimization up to the scheduler.

This model increases the complexity of the CTC scheduler, but keeps the TMC software model straightforward and manageable. CTC schedulers built in anticipation of this architecture are able to handle the multi-robot issue introduced by hybrid cluster tool configurations.

Conclusions

The hybrid cluster tool architecture combines an atmospheric front-end buffer station and a vacuum-based cluster tool. It offers equipment suppliers and fab integrators a flexible, cost-effective factory interface designed to meet 300mm fab requirements for tool WIP buffer stations, automated FOUP exchange, and increased tool utilization.

The atmospheric buffer station can become a common physical interface between tool and fab, irrespective of whether the processing tool is a cluster or batch tool. For batch process tools that do not require a vacuum transport system, an atmospheric buffer station with robotic wafer handler operating as a factory interface is an attractive alternative. Fabs can specify buffer cassette station requirements based on optimal factory scheduling versus cost constraints.

This new hybrid architecture creates new challenges for control software developers which are best resolved by contemplating the use of a single authority (the CTC) to coordinate both the atmospheric and vacuum transport subsystems. To maximize overall system efficiency and ensure effective utilization of the two transport components, the cluster tool controller (CTC) must be fully exposed to the hybrid architecture.

Dan Camporese is Chief Scientist at Brooks Automation Software Corp. currently focusing on scheduling, advanced process control, and CIM integration technologies. He has a Ph.D. in Electrical Engineering from the University of British Columbia, Canada, where he served as an assistant professor for four years teaching Computer Architecture and VLSI design. Working with Brooks Software (formerly Techware Systems Corporation) from 1985-1986 and 1990-present, Dan has been instrumental in developing their natural language approach to real-time equipment control. In his former capacity as R&D Manager, he supervised development projects for MIMO control, MESC-conformant equipment control, graphical user interfaces, and sensor bus I/O. He has conducted research in fine-grained multiprocessing, neural network hardware implementation, VLSI CAD and the application of neural networks to real-time closed loop control.

For more information: Brooks Automation Software Corp., 100-13777 Commerce Parkway, Richmond, B.C., V6V 2X3, Canada. Tel: 604-214-5000, fax: 604-214-5001.