Years to End of Life: How Reliable is EOL Forecasting?
The Need for Proactive Obsolescence Management
Electronic component obsolescence presents challenges to product design that are often costly. Managing the risk associated with component lifecycles and weighing whether or not parts will last the whole lifecycle of a product is no easy task. As an engineer, you must balance product quality and functionality with specifications, regulation requirements AND an obsolescence plan that factors both immediate and long term component end of life.
In the event of part obsolescence, companies typically fall victim to a high demand and limited market availability situations.
Our 2015 PCN report found that 20% of product change notices (PCNs) were for parts with last time buy dates of “immediately”; meaning that waiting for a PCN may result in a costly re-design.
The solution? End of life forecasting. End of life forecasting data provides companies with the insight necessary to manage risk during the initial component selection process. This gives engineers the ability to match their component lifecycles with that of their projected product lifecycle and plan ahead with alternate parts that have similar form fit and function.
Forecast Years to End of Life
SiliconExpert provides estimated Years to End of Life (Y-to-EOL) as a data point in our electronic component database. The algorithm was developed as a joint partnership between SiliconExpert and The Center for Advanced Lifecycle Engineering (CALCE) at the University of Maryland to help customers plan for and mitigate part obsolescence. The algorithm looks at historical patterns to determine the procurement life of an obsolete part and estimate the Y-to-EOL for active components.
Every two years, SiliconExpert measures the accuracy and reliability of our End of life algorithm. Below is the latest report using 6 years of data.
To measure the Y-to-EOL reliability, parts classified under SiliconExpert’s semiconductor taxonomy with an “Active” Lifecycle status on 1/1/2009 were selected. This resulted in a sample size of approximately 1 million parts.
The Y-to-EOL of these components was computed with a prediction date range of 1/1/2009 to 1/1/2015, diving the parts into four categories:
Components accurately predicted to be EOL: The current lifecycle of the part is obsolete and the algorithm predicted it would be obsolete within 6 years.Components inaccurately predicted to be EOL: The current lifecycle of the part is active and the algorithm predicted it would be obsolete within 6 years.Components accurately predicted to be active: The current lifecycle of the part is active and the algorithm predicted it would be active in 6 years.Components inaccurately predicted to be active: The current lifecycle of the part is not active and it was predicted to be active.
A similar analysis was conducted in 2013 and the results between 2015 are compared to these results to determine if the reliability percentage has improved over time.
The overall accuracy of the SiliconExpert Y-to-EOL algorithm for 2015 is 89% an increase from the 2013 rate of 85% as highlighted in Figure 1. The overall inaccuracy percentage dropped from 15% in 2013 to 11% in 2015.
Between 2013 and 2015 the overall accuracy improved for both accurate and inaccurate predictions. These changes are due to additional parts being added to the SiliconExpert database and a modification to the Y-to-EOL algorithm. SiliconExpert is constantly adding additional parts to the database. The more parts in the database the better the algorithm becomes in determining historical trends, especially if the parts being added are obsolete.
Over the past two years SiliconExpert has enhanced the Y-to-EOL algorithm to make it more dynamic and more reliable by adding additional factors to the algorithm. We will continue to monitor the reliability of the algorithms and make tweaks and changes as necessary to ensure the Y-to-EOL algorithm provide our customers with confidence as they mitigate risk associated with reactive obsolescence management.
Test results are outlined in more detail in two section below.
Predictions for Parts Classified as EOL in 2015
Figure 2 shows the accuracy percentages for parts classified as active on 1/1/2009 and obsolete on 1/1/2015. A higher percentage is good for accurate predictions, a lower percentage is better for inaccurate predictions.
An accurate prediction for an EOL part indicates that on 1/1/2009 SiliconExpert predicted the part would be obsolete within 6 years and it is. In 2015 the accuracy percentage increased +3% over the 2013 value to 87%.
Inaccurate predictions for an EOL part is based on the SiliconExpert prediction on 1/1/2009 that the part would still be active in 6 years but the currently lifecycle status is obsolete. In 2015 the inaccuracy percentage dropped to 13%, a drop from 16% in 2013.
Predictions for parts classified as Active in 2015
Figure 3 shows the accuracy percentages for parts classified as active on 1/1/2009 and active on 1/1/2015.
An accurate prediction for an active part is when SiliconExpert predicted on the part would be active on 1/1/2015 and it is. The accuracy prediction increased to 92% in 2015, compared to 85% in 2013.
Inaccurate predictions for active parts are when the prediction on 1/1/2009 stated that part should be obsolete by 1/1/2015 and the part is still active. The inaccuracy prediction dropped to 8% in 2015 compared to 15% in 2013.
Accurately Forecast End of Life
You don’t want to design a new product only to find out that a critical component becomes obsolete a week after launch. Long product lifecycles mean high sensitivity to lifecycle changes and a real need for a proactive approach to obsolescence management. End of life forecasting data provides companies with the insight necessary to manage risk during the initial component selection process. However, a predictive model that is inaccurate can potentially add risk. This paper provides transparency into the Y-to-EOL algorithm used by SiliconExpert enabling customers to have increased confidence in their obsolescence management forecasting.
FEBRUARY 5, 2016 BY AMANDA M (EDIT)3