標題: 藉由掃描鏈重序達到低耗能嵌入測試之設計
Low Power DFT by Scan Chain Reordering
作者: 韓開旭
Kevin Kai-Shu Han
單智君
Jyh-Jiun Shann
資訊科學與工程研究所
關鍵字: 低耗能;掃描鏈;嵌入式測試;重序;旅行商人;low power;scan chain;DFT;reordering;traveling salesman
公開日期: 2001
摘要: 循序電路(Sequential Circuit)的測試往往藉由插入掃描鏈(Scan Chain)的方式將測試資料(Test Pattern)傳送給待測的電路(Circuit Under Test, CUT); 藉由掃描鏈串聯的傳輸方式,可大幅減少傳輸測試資料時所需的資料埠(Data Ports)。可是當待測電路越來越複雜,為達到要求的偵錯率(Fault Coverage),所需的掃描鏈長度便相對的增加。此時,測試資料的傳遞不僅耗時,而且其耗能很可能超過晶片設計時所容許的最大規格,進而造成測試的誤失,甚至有可能因為瞬間的大電流造成待測電路的毀損!因此一個經過省能考慮並顧及最大繞線距離的掃描鏈是相當重要而且實用的設計需求。 在本篇論文中,我們嘗試將掃描鏈定序(Scan Chain Ordering)轉化成一個「旅行商人」(Traveling Salesman Problem)的問題模式,其中包括了全連通圖(Fully Connected Graph)的建立與權重(Weight)的計算,然後提出一個「最小權重優先」(Lowest Weight First)演算法以找出新的定序方式。模擬結果顯示,重新定序後的掃描鏈其資料傳輸時的總耗能最多可節省33.8%,最大功率消耗則最多可節省29.8%。另外,將本篇論文的方法應用在掃描鏈繞線長度上的縮減時,重新定序的掃描鏈其繞線總長度最多將可縮減至原來的5%以下。 此外,本論文提出的方法在執行時間方面非常有效率,相較於其他掃描鏈重序的方法,我們的方法所需的計算時間僅為他法的千分之一至萬分之一不等。此將更有助應用於超大型積體電路測試時的實際情況與需求。
Scan-Chain insertion is widely used in design for testability (DFT) technique to test sequential circuit. The test pattern could propagate to the Circuit Under Test (CUT) through the scan chain serially to reduce needed data ports. However, the power of scan testing is usually three to four times larger than the power under normally functional operation of the same circuit. The large power of testing may fail the testing result and even damage the circuit under test (CUT). So that an efficient method to reduce power consumption and total routing length concurrently is important for nowadays IC design. In this thesis, we propose a method to lower down the power consumption by scan chain reordering. The method also could take the total routing length of the scan chain into consideration concurrently. By modeling the whole system as a traveling salesman problem (TSP) and solving it using a Lowest Weight First (LWF) heuristic algorithm, the saved energy of the reordered scan chain could up to 33.8% and the saved maximum power consumption could up to 29.8%. Moreover, the total routing length could also be minimized to be shorter than 5% of the original one. The time efficiency of our proposed method is better than the previous approaches for about three to four magnitude of order. That makes our proposed method more feasible and suitable than other approaches to deal with nowadays’ more and more complicated IC designs.
URI: http://140.113.39.130/cdrfb3/record/nctu/#NT900392100
http://hdl.handle.net/11536/68508
Appears in Collections:Thesis