基於類神經網絡平台利用製程監測反應晶片狀態

Abstract

本篇論文提出了以類神經網絡為主的架構來找出晶片狀況。隨著製程的演進，製程飄移的情況越來越嚴重。製程飄移甚至在良率上佔有一席地位。然而，製程飄移的量測需要大量的花費，其中包含額外的測試架構以及特殊的測試機台。另一方面，以環形振蕩器為基礎的製程監測在業界廣為採用在反應晶片速度。以環形振蕩器為基礎的製程監測採用了不同的振盪器架構將量測到的速度分解成不同的製程飄移參數。我們提出的架構根據實際量測出來的速度配合模擬資料來尋找最有可能的晶片狀況。整個實驗環境是採用聯電28奈米製程。

This paper proposes a neural network based framework to identify circuit condition where the ring oscillator based process monitor is. As the process shrinks, the variation becomes serious issue on product manufacturing. Moreover, the variation of process plays an important role on the yield. But the measurement on the variation of the process requires huge test cost on additional test structure and auto test equipment (ATE). In the other hand, ring oscillator based process monitor is widely used in industry design to sense the chip performance. Ring oscillator based process monitor uses different structures to decompose the observed process monitor speed into multiple variation parameters. The proposed framework took the measurement to identify the variation. We applies neural network as the spice model to search the closest condition of the location of process monitor. The experimental result is based on the UMC 28nm technology.