The semiconductor industry is highly competitive and cyclical. In order to gain an edge over competition, constant innovation and the ability to adapt quickly, are very important. While on the one hand, semiconductor manufacturers are investing heavily in research and development; on the other, there is also a race to improve efficiency in design, manufacturing and operations. Building an integrated, enterprise-wide data warehouse provides a good way to leverage the organization’s data assets and increase operational efficiency.
The success of any manufacturing enterprise depends on the efficient execution of all its business functions like Manufacturing, Procurement, Order Management, Supply Chain Planning, Finance, Sales & Marketing, Product Quality, Logistics & Shipping, etc. A huge amount of data is generated as these functions operate. Also, these business functions rarely operate in isolation and there is significant interdependence between them.
In the semiconductor industry, for example, for any new sales order created, a demand signal is created to the manufacturing entities (factories incl. foundries and subcons), based on which the supply chain and manufacturing operations can be planned. In the Finance system, it is represented as an order book entry, which is tracked to revenue. The revenue and order backlog history is in-turn required for the Sales & Marketing team to identify a customer’s buying history, and identify more efficient pricing models. The manufacturing operation itself generates data, which is critical for controlling product quality and plant efficiency.
Setting up an Enterprise Data Warehouse brings together all this manufacturing and operational data at one place, providing a single version of truth. Most semiconductor manufacturers are multi-billion dollar entities, with manufacturing plants spread globally. This makes it even more important to consolidate the data into a one-stop-shop for reporting and analytics, so that uniform business rules are applied.
A recommended approach would be to capture all the detailed data generated from the manufacturing processes during Fab, Probe, Assembly and Test phases, into a Hadoop data lake. The data can then be summarized to the appropriate level and made available in the EDW for reporting. The operational data (Finance, Sales, Procurement, etc.) is typically more structured as it resides in ERP systems. It can be brought in directly to the data warehouse. Once the data is in the warehouse, it becomes possible to build cross-functional analytics to gain business insights and drive operational decisions.
In order to deliver the value of the EDW, it is recommended to create a single layer or Master data (e.g. products, customers, vendors, plants, locations, etc.). This Master data ensures uniformity in reporting, which is especially important for an industry operating through geographically distributed plants. Consolidating the data this way, enforces a disciplined approach to data governance. Even with the transactional data, it should be modelled to have Common Functional Names that are identifiable and understandable by all business users. This encourages a collaborative and democratic approach to data exploration and minimizes dependence on IT, delivering business insights when they are required. Of course, the power of self-service data visualization tools makes this even simpler.
The end-goal of this exercise would be to gather all the data in EDW to the appropriate level of granularity and then apply “functional masks” on this data to make it behave as per the requirement. Consider for instance, that the EDW has lot level inventory and order backlog data gathered from MES and ERP systems in real-time. We might have a reference data with average sales price. Now applying this price information to the inventory data can allow us to do near real-time financial forecasting. This would be a “Finance mask”. Similarly, we could apply masks that compute quality metrics on this lot level data to gain insights into Product Quality.
Let me end with some tangible examples in the Sales and Marketing function to illustrate how an integrated EDW readily translates to business insights:
Pricing Leakage Avoidance – Semiconductor manufacturers enter into annual agreements with their major OEM customers to offer most competitive price. While doing so, they need to have the insights into the history of direct sales, distributor resales and forecasted sales to gauge past compliance and optimize the prices. Leveraging this intelligence can greatly improve the efficiency of the negotiation process and provide savings in order of millions.
Distributor Reserves Requirement – Anywhere between 35% and 50% by volume of semiconductor sales happens through distributors, including e-tailers. Semiconductor companies need to reserve some portion of their revenue, as pay back to distributors to protect against variation in demand and fluctuations in prices. Here again, we need to marry the data from Finance, Marketing and Pricing systems to optimize this process.
Statistical Pricing – In order to provide competitive pricing to the customers, semiconductor companies can lookback into customer’s historical sales and revenue records spanning across products and regions. By applying classification and scoring algorithms, it is possible to predict a target, floor and ceiling price value for any transaction, such that the customer gets an attractive price for a volume purchase and the manufacturer is safe-guarded from underquoting.
The possibilities are indeed limitless.