GHz瞬态红外下降监测器：片上环形振荡器应用

66 浏览量更新于2024-07-14 收藏 11.27MB PDF 举报

"本文提出了一种基于片上环形振荡器的GHz瞬态红外下降监测器（TRO），用于解决现代集成电路设计中的功率完整性问题。随着硅技术的不断缩放，高速GHz操作频率下的开关活动加剧，导致了IR-drop噪声的显著增加，这可能引发时序故障、异常复位和SRAM翻转等功能性失败。因此，对IR-drop进行实地监控变得至关重要。然而，直接测量瞬态IR-drop波形通常需要高昂的设计或设备成本。TRO由全数字元件构成，可以轻松集成到现有的IC设计流程中，几乎不增加额外开销。与传统的瞬态IR-drop监测器不同，TRO通过测量IR-drop波形的宽度和平均电流来评估IR-drop的影响。" 文章详细探讨了在集成电路设计中，由于技术进步和高速操作，IR-drop（电压降）已成为一个日益严重的问题。IR-drop是指电流流过电路时，由于电阻引起的电压损失，可能导致性能下降甚至系统故障。现有的IR-drop监测方法成本高且实施复杂，而TRO（Transient Ring Oscillator-based IR-Drop Monitor）提供了一种新的解决方案。 TRO设计的核心是利用环形振荡器，这是一种常见的数字电路，由多个反相器组成，形成一个自反馈回路。在电压降的影响下，振荡器的频率会改变，因为电压下降会改变晶体管的工作速度。通过监测振荡器频率的变化，TRO能够间接测量IR-drop的幅度和持续时间。这种监测方法的优势在于其全数字化特性，使得它可以直接嵌入现有的集成电路设计中，无需额外的硬件或昂贵的测试设备。此外，TRO关注的是IR-drop波形的宽度，这反映了电压恢复的时间，以及平均电流，它能反映出系统功耗的动态变化。通过分析这些参数，设计者可以更好地理解和预测IR-drop对系统性能的影响，从而在设计阶段就采取措施减轻IR-drop问题。论文进一步可能涉及了TRO的实现细节，包括振荡器的结构优化、信号处理算法以及如何将监测结果与仿真结果对比以验证其准确性。此外，实验部分可能会展示TRO在不同工作条件下的性能，如不同负载条件、温度变化和电源电压波动下的响应，以证明其在实际应用中的有效性。 TRO为实时监测和管理集成电路中的IR-drop提供了一种成本效益高且易于实施的方法，对于推动高性能、低功耗的集成电路设计具有重要意义。这一技术对于集成电路设计者、半导体制造企业和电子设备制造商来说，都是一种有价值的工具，有助于确保产品的稳定性和可靠性。

IEEE Proof

6 IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS

Hence, the (∂T/∂VDD) of a working Fast RO is the

function of temperature K.

4) The K-(∂T/∂VDD) relationship should be built during

production test. Therefore, in-ﬁeld (∂T/∂VDD) can be

determined after obtained the in-ﬁeld temperature K.

5) With in-ﬁeld (∂T/∂VDD) determined, the N

norm

-IR

width-IR peak relationship is ﬁxed. Then IR peak can

be decided by the measured N

norm

and IR-drop width.

6) Since process variations and temperature inﬂuence

the N

norm

-IR width-IR peak relationship by affecting

(∂T/∂VDD). Thus a set of N

norm

-IR width-IR peak

lookup tables can be prepared in design stage

through SPICE simulation at various (∂T/∂VDD).

After in-ﬁeld (∂T/∂VDD) is determined, a proper

norm

-IR width-IR peak lookup table can be picked out.

In summary, during in-ﬁeld IR-drop monitoring, only tem-

perature K (represented by noise free Fast RO oscillation num-

ber N

), N and IR-drop waveform width need to be measured,

which guarantees TRO’s short measurement time. During

data analysis or customer return, the proper (∂T/∂VDD)

of a speciﬁc Fast RO can be determined by K. Then the

proper N

norm

-IR width-IR peak lookup table can be selected

from simulated lookup table set. Finally, the IR peak can be

obtained.

C. Analyses of TRO’s Performance Under Process and

Temperature Variations

The process variations and temperature affect measurement

accuracy, and should be minimized. TRO uses three elements

to reconstruct the transient IR-drop waveform off-chip: 1) the

IR-drop width; 2) average; and 3) the N

norm

-IR width-IR peak

lookup table correlating IR-drop peak to the width and aver-

age. The process variations and temperature’s impacts on the

above three elements are controlled and minimized as follows.

1) The IR-drop width is obtained by the Edge Detector,

and calculated according to [(j− i) + 1]W, where i and j

are the ﬁrst and last ﬂipping branch index and W is the

sampling window length. Process variations and tem-

perature inﬂuence the value of i and j by affecting the

Edge Detector branch’s IR-drop sensitivity. For exam-

ple, if a strong sub-branch is too much longer than its

pairing weak subbranch, due to large process or temper-

ature variations, the branch signature cannot ﬂip from

0 to 1 even with large IR-drop. To make the sensitivity

stable, the branches are automatically calibrated, during

which the length of the strong subbranch is increased

one to four big buffer delay (t

) until the branch sig-

nature ﬂips from 1 to 0. Thus, after calibration, the

delay of strong subbranch is 0-t

longer than that of

weak subbranch. For advanced technologies, t

can eas-

ily be in the range of 10–15 ps. As mentioned above,

the calibration process ﬁnishes within 16 clock cycles. It

can be considered that the temperature keeps the same

through measurement, and the length of sub-branches

keeps the same through calibration and measurement.

The branches’ sensitivity after calibration can be deter-

mined by keep reducing the global VDD to (VDD−S),

Fig. 10. Distribution of branch sensitivity before and after calibration. It can

be seen that the variance of sensitivity shrunk 73.7% after calibration.

until the branch signature ﬂips to 1. Then S is the

branch sensitivity. Fig. 10 shows the distribution of S

under 25% V

, 10% L,5%W, and 1% t

process

variations across −40

◦

C–100

◦

C before and after cal-

ibration. It can be seen that calibration improves the

stability of branch sensitivity by 73.7%. Also according

to the IR-drop width equation [(j−i) + 1]W, the process

and temperature variations also affect the measurement

result by affecting W, in other words, changing the delay

of BUF

to BUF

[M/W−1]

in Fig. 3. However, by selecting

the buffer (composed by two inverters) of the same type

as Fast RO with low delay variation, the actual silicon

W at measurement temperature can be obtained from the

recorded N

by M/nN

, where N

is the noise-free Fast

RO oscillation number, M is the measurement window,

and n is the inverter number of Fast RO.

2) The average IR-drop is represented by the normalized

counter value N

norm

. As the counter counts the rising

edge of Fast RO oscillation waveform, the process vari-

ations and temperature’s impact on the counter behavior

is negligible.

3) The last element is the lookup table correlating IR-drop

peak to the width and average. From (8), it can be seen

that, under given IR-drop waveform V(t) and measure-

ment window M, the process and temperature variations

only affect the parameter of ∂T/∂VDD. Fortunately,

the in-ﬁeld ∂T/∂VDD can be determined by recording

the in-ﬁeld temperature and checking the K-∂T/∂VDD

lookup table generated during manufacture test or cus-

tomer return. Therefore, process variation and temper-

ature’s impact on the N

norm

-IR width-IR peak lookup

table can be eliminated.

In summary, the process and temperature variations’ impacts

are minimized through branch calibration, cell selection, and

temperature rectiﬁcation.

III. F

LOW FOR TRO IMPLEMENTATION,

EASUREMENT, AND ADAPTATION

The overall implementation, measurement, and adaptation

ﬂow of TRO system is shown in Fig. 11. Since the TRO system

剩余27页未读，继续阅读

weixin_38672815

粉丝: 11
资源: 869

GHz瞬态红外下降监测器：片上环形振荡器应用

tro:电子元件协同数据库

Mice-tro:Mice-tro是一个网络应用程序，它将允许用户在线创作音乐

ParisM-tro:学校项目 2013 年 12 月

Quan_ly_phong_tro:Phầnmềmquảnlýnhàtrọ

tro头

PWA01TRO

PWA00TRO

ICS3U-In-tro-Comp-Sci：计算机科学概论

tro_dnevni_mod

Phong-Tro-DN

最新资源