Photoresist Usage Forecasting and Stocking

The most expensive chemical subset used in semiconductor manufacturing consists of photoresists, antireflective coatings (ARCs) and polyimides. Accurate usage forecasting of these specific photochemicals is essential for a consistent operation of the semiconductor fab. Inventory overstocking can result in expensive waste, and understocking can result in the need for expedited production and shipping from the chemical supplier, with possible halts in production.

Most advanced photoresists have a short shelf life — typically six months — often requiring cold storage conditions, thus careful inventory planning is critical. Fabs with many products covering multiple technology nodes typically have a high number of different photoresists, making accurate usage forecasting a challenge.

Consignment stocking of expensive photoresists, in which a large inventory supply is delivered to the semiconductor fab and only paid for as consumed by manufacturing, offers benefits (and some risks) to both the supplier and customer. Consignment stocking of photoresists can only be successful with accurate usage forecasting to allow for adequate raw material planning and production scheduling by the photoresist supplier.

The simplest method to project photoresist usage is to use historical unit usage rates — typically an average of the previous three months — to smooth out chemical delivery variations. This method is effective only when the semiconductor fab is running at a steady production rate. However, the semiconductor business is notorious for being cyclical in nature with large swings in product demand, greatly complicating accurate photoresist usage forecasting.

A more accurate methodology for calculating photoresist usage takes into account all of the typical variables of wafer production, and allows for implementation of a successful consignment stocking program.

Photoresists (which includes ARCs and polyimides for the purpose of this article) are applied to each wafer with a specific quantity, which is called the coating volume. The coating volume is very small, often <1.0 mL/200 mm wafer with use of a solvent pre-wet. The resist usage rate can be calculated by multiplying the coating volume times the number of wafers run per day, and dividing by 1000 to convert to liters. The volume in liters per day is multiplied times 30 for monthly usage, and then divided by the package volume to obtain units per month. There are a number of variables that can affect the accuracy of the calculated monthly usage involving both the photoresist coating volume and the number of wafers processed.

The amount of photoresist dispensed onto the wafer can vary from apply track to track depending on apply pump accuracy, and must be physically measured and then periodically checked for consistency. If a solvent pre-wet is used to reduce the photoresist coating volume, the solvent volume should also be monitored for consistency. Occasionally, the coating volume is increased to resolve defect or coating quality issues and then not returned to the original volume, causing an unexpected increase in photoresist usage.

The newer phototrack dispense pumps are very accurate and typically only require dispense volume calibration on initial installation in the fab. Tracking photoresist consumption by phototrack can reveal usage anomalies that can be investigated and corrected.

The number of wafers run per day of a particular semiconductor product is projected by a production planning group and adjusted each month. The product specific “wafer starts per day” figures form the basis of calculating photoresist usage out in time. It is important to take into account the one to two month time lag between starting a wafer in production and actual photoresist application.

For accuracy, the number of product wafers run per day using a specific photoresist needs to include all monitor wafers in the total count since they are also coated with photoresist. Monitor wafers are most commonly used to measure photoresist film thickness and coating defect levels. The average percent rework must also be added to the total.

Resolving wafer defects often requires purging photoresist to drain and running multiple blank monitor wafers to eliminate the defects. Changing point-of-use filters also involves purging photoresist to drain trapped air bubbles. Another significant factor in photoresist usage is the need for periodically dispensing photoresist to the waste drain to prevent drying of the dispense tips. The frequency of the periodic dispense can have a large impact on low-usage photoresists, where it is possible to dispense more photoresist to drain than is coated on product wafers. This is a problem with new product programs that are under process development or slowly ramping in production.

The number of tracks loaded with a particular photoresist also has an impact on waste levels if the “deployment” is excessive and does not match wafer production volumes, resulting in a higher impact of periodic dispensing to waste. A frequent evaluation of the photoresist track deployment compared with wafer production is needed to correct imbalances.

The number of photoresist bottles consumed per track over a 10-month time period demonstrates the wide range of usage, with the highest number of tracks running a particular photoresist compared with a small usage variation with the lowest number of tracks (Figure ). These results indicate that the periodic dispense waste will have a greater impact on the photoresist loaded on a large number of tracks.

The number of photoresist bottles consumed per track over a 10-month time period.

All of these “waste” factors — resolving wafer defects, changing filters, periodic dispense to drain and overdeployment — can result in as much as 20% higher usage of a particular photoresist. The percent of waste photoresist can be determined by comparing the calculated usage rate with the actual three-month average unit consumption.

Taking all of these factors into account will greatly improve usage forecasting accuracy, and will prevent both over and understocking of these expensive chemicals, allowing for the full benefit of consignment stocking.

In consignment stocking, the supplier ships inventory, typically a three-month supply, to the customer's chemical warehouse, and the customer pays for the product as it is consumed. To maximize photoresist shelf life, a supplier “build to order” process is used in place of “building ahead” photoresist batches in anticipation of orders.

While the supplier still “owns” the chemical inventory, the customer is fully responsible for maintaining proper storage conditions, which includes refrigeration. The supplier is not responsible if the chemical shelf life is exceeded during customer storage, but has to provide freshly made material under the consignment agreement to maximize the shelf life, which ranges from six to 12 months.

The customer then provides the photoresist supplier a “monthly inventory pull report” that lists the quantities of each unique photoresist part number removed from the chemical warehouse inventory for use in manufacturing. The report should also include the quantities of remaining inventory. The supplier then invoices the customer for the total amount of photoresists “pulled” from inventory.

Most photoresist suppliers cite the time and manpower required for quality control (QC) testing as a major factor in the variable cost of their products. The consignment “build to order” process allows the supplier to make single, large batches of each chemical to meet quarterly deliveries to a semiconductor customer, reducing overall QC testing on a yearly basis.

Maintaining a large photoresist inventory at all times — from one to four months supply — will greatly reduce unanticipated “emergency” orders caused by chemical shortages.

The consignment “build to order” process has the dual cost benefit of eliminating supplier stocking inventory in the warehouse in anticipation of customer orders and then, once the chemical orders are built, shipping the entire quantity to the customer for storage. The contracted consignment order leadtime ranges from four to eight weeks, depending on the supplier's production location, allowing the supplier sufficient time for production planning and raw material ordering.

In addition to the obvious advantage of delayed payment to the supplier with consignment stocking, maintaining a large inventory of photoresists can usually eliminate the need for emergency orders caused by unanticipated shortages. Transportation costs in general are reduced because of quarterly instead of monthly photoresist deliveries.

The large consignment inventory held by the customer also protects the fab's production due to chemical delivery disruptions as a result of severe weather, natural disasters and transportation delays.

The major risk to the supplier is a customer overprojecting usage or deciding to no longer use a product, resulting in unneeded raw materials with associated shelf-life expiration. The supplier may have to scrap this inventory if another customer is not available or it cannot be “reworked” into another product. Delayed payments by the customer to the supplier could cause cash flow problems if not properly planned for.

Maintaining a greater than one month supply of photochemicals with short shelf life runs the risk of shelf-life expiration if fab production slows down. The customer has to maintain proper storage conditions of consigned inventory and practice a strict “first in, first out” delivery process to ensure the oldest photoresist lots are delivered to manufacturing ahead of new lots to maximize the chemical's remaining shelf life. The customer's chemical warehouse must maintain cold storage, which includes both refrigeration and freezer facilities, and have emergency backup capability if there is a power outage.