電路老化監測系統及其方法

專利類型

發明

專利國別 (專利申請國家)

中華民國

專利申請案號

109138488

專利證號

I 768532

專利獲證名稱

電路老化監測系統及其方法

專利所屬機關 (申請機關)

國立彰化師範大學

獲證日期

2022/06/21

技術說明

[中文摘要] 本發明中單一老化因子強化影子樹(Single-Aging Factor-Enhanced Shadow-Tree)部份，針對系統晶片電路各單位進行模擬(通常為Verilog-A/HSPICE)評估老化傾向，再對整體電路編譯解析(parsing)，挑選出面積與侵入性最低且具有單一老化因子老化傾向最高的子電路進行就地複製，使具有相同輸入、狀態和歷史，再以增加電壓以加速老化；因為常為數個元件構成之樹狀子電路，故稱為影子樹(Shadow Tree)。 單一老化因子強化影子樹(Single-Aging Factor-Enhanced Shadow Trees, SAFEST)單獨運行監測時，係以其單一老化因子進行老化成因判讀，以進行適當緊急措施。 另一方面，一則因為SAFEST靠近正常電路很近，為避免侵入性(Invasibility)，加速老化之電壓不可太高；二則，因老化癥狀(Aging Symptom)可能多種重疊顯現為癥候群(Syndrome)，其成因也可能由多種因子(Aging Factor)歷程在激化(Activator)催化(Catalyst)下，才呈現出可量測之響應，所以，勢必存在部份特徵(Features)必須以類神經網路(Neural Network)儘早識別；再則，老化初期SAFEST響應不高，重疊性大，必須有合適之分類註標以進行監督學習。因此，我們在第二部份發明了一不同電壓、不同週期以及不同老化型態犧牲電路的單一老化因子強化震盪環(Single-Aging Factor-Enhanced Rings, SAFER)，其最高速老化元件或眾數元件之老化型態將在第一階段學習時用以當作分類器的學習註標(Annotation, Label, Tag)。 在第二階段，分類器已取得理論上合理之準確度(Accuracy)時，將用以挑選同一老化型態之SAFER的電壓和週期，使其老化程度為初期，加入監測行列，用以提升監測準確度。 [Abstract] To provide high security, self-driving, and a lot of application utility and multimedia, a high-speed, high volume and high-reliability SoC should be required for modern automotive electronics. Usually multi-core CPU, high-speed GPU and /or TPU and IOP, and even reconfigurable FPGA are designed in the SoC. The set of electronics include the SoC will take the responsibility of the automotive mission in such a harsh and noisy environment, the life time have to be long enough and predictable. At least we have to take some conservative emergency measures, for example, stopping unnecessary utilities, starting up the cooling system, degrading the executing frequency, and then warning for changing to manual driving, and even pulling over and waiting for support。Meanwhile the failure prediction is the conservative but most critical function, and for this significant function, we then propose this patent. This patent proposes a failure-predicting aging monitor and a method thereof. The monitor consists of three major parts. The first part is a set of Single-Aging Factor-Enhanced Shadow-Trees (SAFEST), which are the duplicates of a Subcircuit with the same inputs and tending to be aged by a single aging factor, including NBTI, HCI, fatigue and TDDB, and so on. The second part is a neural-network-based classifier, which can be easily implemented by either hardware and/or software. The last part is an array of the Single-Aging Factor-Enhanced Rings (SAFER), which are ageing-accelerated with a higher voltage than the SAFEST. A selector of the SAFERs caused by the same factor is employed for choosing the highest response as the label (data annotation) for supervised learning of the neural network in the first stage. In the second stage, the trained neural network will be utilized to select proper SAFERs for each aging factor. They will join the classifier for improving the perception accuracy in the third stage.

備註

連絡單位 (專責單位/部門名稱)

研究發展處

連絡電話

04-7232105轉1858

網址