Method and circuit arrangement for determining power supply noise

[0001] This is the national stage filing under 35 U.S.C. §371 of International Application No. PCT/IB2004/050107 filed on Feb. 12, 2004; which claims foreign priority benefit under 35 U.S.C. §119 of European Application No. 03100444.3 filed on Feb. 25, 2003.

[0002] The present invention relates to a method and circuit arrangement for determining power supply noise of a power distribution network, in particular a circuit arrangement provided on an integrated circuit comprising the power distribution network.

[0003] Especially in high-speed digital integrated circuits (ICs) the problem of power supply noise (PSN) is faced. Due to their complexity, integration density and speed of operation such modern ICs suffer from signal crosstalk between connection lines. PSN corresponds to variations of the power supply voltage or ground potential caused by the circuit switching activity. The increase in complexity and in integration density leads to an increased electrical interaction between wires of the connection lines distributing supply voltages and active circuits. As a consequence, current spikes produced during switching activities are transformed into voltage bounces or variations in the supply lines. This causes transient reductions in the supply voltage and thus decreases the gate drive strength, which leads to an increased gate delay and a lowered circuit performance. This problem is even worse at higher clock frequencies and lower supply voltages used in current and future technologies.

[0004] PSN is usually managed by a combination of techniques for PSN reduction, e.g. by including on-chip decoupling capacitors or sizing of power supply and ground distribution wires. Moreover, isolation of more sensitive circuits from noisy circuits is a well-known technique. As an example, document EP 0 771 073 A2 discloses a method and apparatus for controlling ground potential variations, i.e. ground bounce, in integrated circuits. Two separate techniques are used simultaneously to diminish the problem of such ground bounce. First, an impedance is placed between a power source distribution line system, i.e. power source bus, on the integrated circuit and an external power source; and between a ground bus on the chip and an external ground. This effectively dampens ground bounce oscillations in the power and ground leads of the chip. Second, capacitance is dynamically added to the pre-drive of an output buffer with a capacitance node. However, according to S. Zhao et al, “Estimation of Switching Noise on Power Supply Lines in Deep Sub-Micron CMOS Circuits”, Proceedings of 13^th International Conference on VLSI Design, Jan. 2000, pp. 168-173, PSN is still a major problem for high-speed systems. Estimation of the worst case switching noise is essential to ensure proper functionality of VLSI (Very Large Scale Integration) circuits. For this reason, on-line monitoring of PSN is an interesting possibility in order to obtain information on the presence of excessive PSN so as to allow a system controller to take actions for correction. Such monitoring should be arranged to allow a location anywhere in the circuit to check the PSN in the local power or ground lines to which the monitoring circuit is connected. Furthermore, the monitoring circuit should be able to check PSN at any given time with a high time resolution and must be sensitive to PSN on supply and ground lines, while the monitoring operations must not be affected by the monitored PSN.

[0005] It is therefore an object of the present invention to provide an improved PSN determination scheme by means of which a flexible real-time PSN monitoring is possible.

[0006] This object is achieved by a circuit arrangement as claimed in claim 1 and by a PSN determination method as claimed in claim 15.

[0007] Accordingly, PSN is measured through its effect on the propagation delay of a delay circuit. Thereby, real-time monitoring of PSN is allowed anywhere within a power distribution network of e.g. an IC. PSN is indirectly measured by means of its effect upon the propagation delay of a dedicated monitoring circuit. The propagation delay is directly dependent upon the supply voltage of the delay circuit, including any and all variations, such as PSN. The less the supply voltage, the greater is the propagation delay and vice versa. Therefore, any variation in the supply voltage due to PSN will be detected, so as to take into account any detrimental effects caused by PSN. This proposed solution allows real-time PSN monitoring, fine grain PSN monitoring, more sensitive and accurate PSN monitoring anywhere within the power distribution network, e.g. on the IC. Furthermore, a plurality of monitoring circuits can be used simultaneously.

[0008] The delay measuring means may comprise a switching means for connecting a current supply to a capacitor, the switching means being controlled by the input signal and the output signal of the measuring means. In particular, the switching means may be controlled by the input and output signals to enable a charging process of the capacitor during the time delay, the output signal being derived from the charge voltage of the capacitor. Thus, a sample and hold measurement technique can be provided, wherein the analog output signal of the sample and hold circuitry, which could be digitized, can then be used to operatively control the concerned signal processing circuit, e.g. the IC, or a part thereof, so as to initiate counter actions.

[0009] The switching means of the above sample and hold circuitry may comprise a first switch controlled by the input signal and a second switch controlled by the output signal. In this case, the first and second switches may be arranged in a series connection between the current supply and the capacitor. Specifically, the first and second switches may comprise transistor switches. Thereby, the whole sample and hold circuitry can be implemented based on a semiconductor circuitry with a capacitor, which allows efficient integration on an IC.

[0010] The measuring means may comprise a buffer circuit for isolating the current supply from the power distribution network. This buffer circuit may comprise switching means and capacitor means, wherein the switching means may be controlled by the input signal so as to temporarily connect the capacitor means to a supply line of the power distribution network. This implementation again simplifies the provision of the buffer circuit as an integrated circuit, and provides an isolation of the measuring process from the PSN.

[0011] Additionally, triggering means may be provided for supplying the input signal to the delay measuring means at a predetermined timing with respect to a clock signal. The triggering means may comprise a plurality of delay lines for delaying the clock signal and selection means for selecting an output of one of the delay lines in response to a selection signal. Such a trigger circuitry provides a mechanism to change the sampling time by means of which the PSN can be sampled at a given time, e.g. when the maximum PSN is produced. By changing the selection, different delays between the global clock signal and the input signal of the measuring means can be generated to search for the maximum PSN detection value.

[0012] The delay circuit may be arranged to delay the input signal by a delay time which is at least ten times smaller than a characteristic time of a waveform of the power supply noise. This allows a proper sampling, since the sampling period of the PSN monitoring process is much smaller then the duration of the PSN. The PSN behavior will thus be collected with enough accuracy.

[0013] Furthermore, an output processing circuit can be provided for processing the output signal to detect an excessive power supply noise. This may be achieved by a comparator for comparing the output signal with a predetermined threshold. This output processing provides the advantage that a convenient digital output is generated which reports if PSN is too high.

[0014] The measuring step may either be performed at one point of the power distribution network or at multiple points of the power distribution network. Furthermore, the measuring step may be performed in one or a plurality of predetermined clock cycles or in each clock cycle of a synchronous integrated circuit comprising the power distribution network.

[0022] The preferred embodiment of the present invention will now be described on the basis of a monitoring circuit for real-time monitoring of PSN of a digital circuit. The monitoring strategy comprises measuring the PSN by its effect on the delay of the monitoring circuitry. The monitoring circuit transforms the variable delay into a variable voltage which can be compared with a threshold or converted later into digital data. The monitoring circuit may thus be used for on-line testing of excessive PSN or as a monitoring part in any scheme that changes the operation circuit parameters to put the circuit within required PSN specifications.

[0023] FIG. 1(a) shows a schematic block diagram indicating an overview of the proposed PSN monitoring circuit according to preferred embodiment. The PSN monitoring circuit comprises a delay block or circuit 10 and a delay measuring circuit 20 which measures the propagation delay between an input terminal and an output terminal of the delay circuit 10. The delay circuit 10 is connected to a power line V_DD and a ground line V_SS of a power distribution network of the digital circuit whose PSN is to be measured. The measuring circuit 20 has its own power supply line V_Y which can be isolated or decoupled from the power distribution network.

[0024] In the preferred embodiment, the delay circuit 10 works as a delay line whose delay value depends on its effective supply voltage. In synchronous systems, the circuit activity and thus also the PSN is triggered by a clock signal used in the digital circuit.

[0025] Accordingly, the input terminal of the delay circuit 10 is supplied with an input signal IN derived from the clock line of the digital circuit. At the output terminal, an output signal OUT is obtained at a delay depending on the propagation delay of the delay circuit 10.

[0026] The measuring circuit 20 comprises switches S_IN and S_OUT which are connected in series between a current source 22 and a capacitor CX. When the input signal IN is at low level or “0” the switch S_IN which is controlled by the value of the input signal IN is open, while the switch S_OUT which is controlled by the output signal OUT is closed when the output signal OUT is at low level or “0”. In this initial situation, the capacitor C_X is discharged and the voltage V_X at the capacitor C_X is zero. The rising edge of the clock signal produces also a rising edge of the input signal IN and the switch S_IN is closed such that a charge current I begins to flow into the capacitor C_X, thereby increasing the Voltage V_X. The charge current I charges the capacitor C_X until the output signal OUT of the delay circuit 10 changes to high level or “1” and opens the switch S_OUT. Hence, the total charge supplied to the capacitor C_X is substantially proportional to the propagation delay of the delay circuit 10. As the propagation delay depends on the effective supply voltage seen by the delay circuit 10, the voltage V_X at the end of the described sampling period depends also on the supply voltage. As the power bounce and ground bounce is produced just after the clock edge, the voltage V_X will be dependent on this power and ground bounce, i.e. the stronger the bounce or variation is, the longer will be the propagation delay and the higher will be the voltage V_X.

[0027] FIG. 1(b) shows signaling diagrams with respective waveforms of the supply voltage V_DD, the ground potential V_SS, the input signal IN, the output signal OUT and the capacitor voltage V_X. When the PSN is zero, the smallest propagation delay t_d0 and the smallest capacitor voltage V_X is observed. At an intermediate PSN value PSN1, an intermediate propagation delay t_d1 and an intermediate capacitor voltage V_X is obtained, while at a higher PSN value PSN2, a higher propagation delay t_d2 and a higher capacitor voltage V_X is obtained. In this way, the capacitor voltage V_X is a sample of the effective supply voltage at the point where the delay circuit 10 of the monitoring circuit is connected. It is noted that in the present example shown in FIG. 1(b) the PSN is generated by a power line bounce and a simultaneous ground line bounce which both lead to a decrease of the effective supply voltage at the delay circuit 10. The highest PSN value PSN2 leads to the indicated change ΔV_X of the capacitor voltage V_X.

[0028] The sampling time corresponds to the moment when the rising edge is produced in the input signal IN which is closely related to the rising edge of the general clock signal of the overall digital circuit. By changing this sampling time, samples of the capacitor voltage V_X can be obtained at different times to thereby check the PSN in the whole clock cycle.

[0029] It is well known that the waveforms of the supply voltage and ground potential are very complex because multiple current spikes of different sizes are observed in and after the clock edge along the whole digital circuit. Moreover, the power and ground lines work as a distributed RLC network and the non-active gates of digital circuit work as a network of lossy capacitors. The current spikes generated on the supply and ground waveforms excite this complex mesh of distributed elements generating complex waveforms that evolve in time and space. The exact waveforms depend on the current distribution, the RLC parameters and the layout of the digital circuit, thus being in general different in each clock period. To deal with this diversity, the PSN monitoring circuit needs to have a time resolution adapted to the PSN waveform,

[0030] To work properly, a sampling period of the PSN monitoring circuit, i.e. the delay of the delay circuit 10 in FIG. 1(a) should be much smaller than the duration of the PSN waveforms. If the characteristic time of the PSN waveform is higher than about ten times the sample period of the PSN monitoring circuit, it will collect the PSN behavior with enough accuracy. The characteristic time of the PSN waveform can be estimated as a characteristic time by knowing the logic deep of the digital circuit, which can be in the range of 20 to 30. Further information regarding the estimation of the characteristic time of the PSN waveform are described in H. Bakoglu, “Circuits, Interconnections and Packaging for VLSI”, Reading, MA Addisson-Wesley, 1990. If the PSN monitoring circuit has a sampling period equivalent to a range from about two to three gate delays, it will be able to capture the interesting features of the PSN waveform. This is the case in the proposed PSN monitoring circuit, such that it can be assumed without great error that the power supply voltage VDD can be regarded constant during the propagation of the input signal IN from the input terminal to the output terminal of the delay circuit 10.

[0031] FIG. 2(a) shows a block diagram of a circuit arrangement comprising a monitoring circuit 100 as shown in FIG. 1 and a preceding trigger circuit 30 for adjusting the timing of the input signal IN. As indicated in FIG. 2, the trigger circuit 30 is connected with its input terminal to a clock line 40 which supplies the clock signal CK, from which the input signal IN is derived. FIG. 2(b) shows waveforms of the clock signal CK, the input signal IN of the monitoring circuit 100 and its output signal OUT, the supply voltage V_DD and the ground potential V_SS. The dotted lines in FIG. 2(b) indicate the start time and stop time of the sampling period, i.e. the propagation delay t_d. If the propagation delay t_d is selected small enough, the effective supply voltage V_DDeff seen by the delay circuit 10 can be regarded constant.

[0032] FIG. 3 shows a schematic circuit diagram of a PSN monitoring circuit according to the preferred embodiment. In FIG. 3, the delay circuit 10 consists of three inverter circuits 12 to which an input signal is supplied via a NOR gate 14 which works as a control gate to disconnect or test the monitoring circuit when desired. Thus, the supply of a clock signal CK to the inverter delay line can be controlled by the logical level applied to a control signal CTRL at the other input of the NOR gate 14. Additionally the clock signal CK is supplied to an inverted gate terminal of a CMOS transistor MP1 which provides a connection between the power supply line V_DD and a buffer capacitor C_Y from which the charging current for the PSN measuring capacitor C_X is supplied. Thus, the transistor MP1 works as a switch which connects and disconnects the buffer capacitor C_Y form the power supplied line V_DD. The buffer capacitor C_Y must be selected large enough to keep its voltage V_Y substantially constant during the sampling process, thus isolating the current supplied to the measuring capacitor C_X from the influence of power supply variations due to the PSN. The serial CMOS transistors MP2 and MP3 correspond to the switches S_IN and S_OUT in FIG. 1(a) and determine the charging time of the measuring capacitor C_X. As indicated in the circuit diagram of FIG. 3, the CMOS transistors MP1 to MP3 are p-channel CMOS transistors.

[0033] Furthermore, an n-channel CMOS transistor MN1 is connected in parallel to the measuring capacitor C_X and is controlled by the input signal IN of the inverter delay line, so as to discharge the measuring capacitor C_X while the input signal IN is at low level or “0”. Thereby, an initial reset of the measuring voltage V_X can be assured.

[0034] The PSN monitoring circuit of FIG. 3 operates as follows. A clock signal CK supplied to the NOR gate 14 closes the CMOS transistor MP1 with its low level state to charge the buffer capacitor C_Y to the power supply voltage V_DD. If the control signal CTRL is at low level, a change of the clock signal CK is transferred to the input of the inverter delay line and sets the input signal IN to a low level which opens the n-channel transistor MN1 and closes the p-channel transistor MP3 to thereby start a charging process of the measuring capacitor C_X. This charging process goes on until the falling edge of the input signal has reached the output of the inverter delay line as a rising edge which closes the upper p-channel transistor MP2 and stops the charging process. The voltage V_X generated at the measuring capacitor C_X then corresponds to the propagation delay of the inverter delay line, which in turn corresponds to the power supply voltage V_DD at the sampling time. The PSN reduces the effective supply voltage applied to the gates of the transistor elements of the inverter delay line and consequently leads to an increased propagation delay.

[0035] The solid line in FIG. 4 shows indicates the relationship between the supply voltage V_DD and the propagation delay t_d as relative values V_DDr and t_ddr for the disclosed chain of three inverters 12.

[0036] This relationship can be expressed by the following equation (1) as defined in Takayasu Sakurai, αPower Law MOSFET Model and its Applications to CMOS Inverter Delay and Other Formulas”, IEEE Journal of Solid State Circuits, April 1990 pp. 584-593:

[0037] $\begin{array}{cc}{t}_d=K\frac{V_{\mathrm{DD}}}{{\left(V_{\mathrm{DD}}-V_{\mathrm{TH}}\right)}^{\alpha }}& \left(1\right)\end{array}$
where K designates a constant of proportionality, V_TH designates the transistor threshold voltage, and α designates a parameter modeling velocity saturation and second order effects. A simulation of the three inverter delay line has led to exemplary values K=1.92×10⁻¹¹ and α=1.37 as the best fit. In FIG. 4, the dashed line follows equation (1) and the solid line has been obtained from a simulation.

[0038] Taking the derivative of the propagation delay t_d with respect to the power supply voltage V_DD, the following expression (2) can be obtained:

[0039] $\begin{array}{cc}\frac{d{t}_d/t_d}{d{V}_{\mathrm{DD}}/V_{\mathrm{DD}}}=\left(1-\frac{\alpha V_{\mathrm{DD}}}{V_{\mathrm{DD}}-V_{\mathrm{TH}}}\right)<0& \left(2\right)\end{array}$

[0040] For the above value of α, and the nominal value of the threshold voltage V_TH, and around the nominal power supply voltage V_DD, this expression is closed to “−1”. This means, that for instance, a 5% decrease in the supply voltage V_DD implies a 5% increase in the propagation delay, which means that the slope of the curve shown in FIG. 4 is practically constant around the nominal value of the supply voltage V_DD. However, if the reduction in the supply voltage V_DD is more severe, the slope increases and linear dependence between the changes in supply voltage V_DD and the changes in the propagation delay is no longer valid.

[0041] On the other hand, the capacitor voltage V_X follows in a first order approximation the following expression:

[0042] $\begin{array}{cc}{V}_x=\frac{I_0{t}_d}{C_X}& \left(3\right)\end{array}$

[0043] Wherein I₀ notes the averaged value of the current I supplied by the switching transistors MP2 and MP3 during the charging period of the capacitor C_X. This averaged current is independent of the changes in the capacitor voltage V_X as long as they are small enough.

[0044] By substituting equation (1) in equation (3) and taking the derivative with respect to the power supply voltage V_DD the following expression (4) is obtained:

[0045] $\begin{array}{cc}\frac{d{V}_X/V_X}{d{V}_{\mathrm{DD}}/V_{\mathrm{DD}}}=\left(1-\frac{\alpha V_{\mathrm{DD}}}{V_{\mathrm{DD}}-V_{\mathrm{TH}}}\right)=\frac{d{t}_d/t_d}{d{V}_{\mathrm{DD}}/V_{\mathrm{DD}}}& \left(4\right)\end{array}$
Which is the same expression as equation (2). This means that for small enough dips or drops in the supply voltage V_DD the change in the capacitor voltage V_X will be proportional to the change in the power supply voltage V_DD.

[0046] The monitoring circuit 100 described above samples the PSN at a given time. However, the exact timing at which the maximum PSN is produced is not known in advance. For this reasons a mechanism to change the sampling time is needed.

[0047] FIG. 5 shows a schematic block diagram indicating the PSN monitoring circuit 100 with the preceding triggering circuit 30 and additional output processing circuit 60. In FIG. 5, a specific example of the trigger circuit 30 is shown, where a plurality of inverter-based delay lines 34 generate different delays between the global clock signal CK and a local clock applied to the PSN monitoring circuit 100. A control or selection signal SEL which can be supplied by a high level controller may be used to select one of the delay lines 34 to investigate the PSN at a given time. The selection of the local clock signals with different delays can be preformed by a multiplexing circuit 32 which is controlled by the selection signal SEL. Thus, the timing of the edge of the local control signal supplied to the PSN monitoring circuit 100 can be shifted based on the selection of the delay lines 34 to a predetermined timing at which a maximum PSN occurs.

[0048] The output of the PSN monitoring circuit 100 corresponds to the analog voltage V_X at the capacitor C_X. In a signal integrity environment it is convenient to obtain a digital output which reports an excessive PSN, i.e. whether the measured PSN is too high. For this reason, a type of analog to digital conversion is required. Due to the fact that the voltage V_X is stored in the capacitor C_X, the acquisition time and the processing time in the output processing circuit 60 can be decoupled.

[0049] If it is required to measure the voltage dip or drop in the supply lines at every clock cycle, then a simple and very fast comparison is needed at the output processing circuit 60. This can be achieved by providing a fast comparator 62 which compares the output voltage V_X with a predetermined output threshold value to obtain a pass or fail signal. In this case, the control signal CTRL must be always at “0” to enable the PSN measuring process at the monitoring circuit 100.

[0050] If the time requirements are not so strict, the PSN monitoring can be preformed only in a given or predetermined clock cycle determined by an activation of the control signal CTRL only during the targeted or desired clock cycle, and keep it at high level or “1” afterwards.

[0051] In this case, the voltage V_X is stored during a relatively long time period in the capacitor C_X and can be processed by a more elaborated and slower but accurate analog-to-digital converter or comparator 62 in the output processing circuit 60. The output threshold value used by the comparator 62 can be set individually based on the requirements for excessive PSN.

[0052] Additionally the output processing circuit 60 comprises a second inverter 64 which is used to reinforce the isolation of the voltage V_X from the digital output, and to provide a positive feedback to the first inverter or comparator 62 in order to avoid the undesirable backward Miller effect produced when the voltage V_X is very near to the comparator threshold. The positive feedback also increases the speed of the comparator 62. Thus, a feeding capacitor C_f may be provided between the analog input and the digital output of the output processing circuit 60.

[0053] In an exemplary implementation of the circuit diagram of FIG. 3, the width of the CMOS transistor has been set to 1 μm, the width of the serial transistors MP2 and MP3 have been set to 0.35 μm, and the n-channel transistor MN1 has been selected with a width of 0.12 μm. Furthermore, the value of the measuring capacitor C_X may be set to 3.7 fF if the comparator 62 is used, while the value may be set to 5 fF if no comparator 62 is used. The buffer capacitor C_Y may have a capacitance of 1 pF. Finally the feeding capacitor C_f may have a capacity of 0.5 fF. It is to be noted that the above figures are only indicated as exemplary figures for a specific integrated circuit and may vary within wide ranges based on the specific application. For the above specific circuit parameters, the sampling period of the PSN monitoring circuit 100 could be set to about 100 ps.

[0054] In summary, the described PSN monitoring circuit 100 is suitable for deep sub-micron technologies and can be located anywhere in the circuit to check the PSN in local power and/or ground lines to which the monitoring circuit is connected. Furthermore, it can be designed as a standard cell for integration purposes. The time resolution can be set high enough to collect the main features of PSN in modern circuits in supply as well as ground lines. The output of the monitoring circuit 100 can be isolated from the PSN, because it uses a locally generated and isolated power supply. The output signal is based on a charge of the capacitor C_X which allows to decouple the sampling time from the processing time of the output. A high-level controller can easily change the sampling time by providing a corresponding controllable trigger circuit. Furthermore, the control input for the control signal CTRL allows the selection of a desired clock circle for PSN measurement. As the monitoring circuit 100 is able to detect on-line the presence of excessive PSN, it can be used in any PSN control scheme to change the operating circuit parameters based on the monitoring output and thereby put the circuit in the PSN specifications.

[0055] It is noted that the present invention is not limited to the above preferred embodiment but can be modified in various ways. For example the arrangement and implementation of the switches S_IN and S_OUT in FIG. 1(a) may be varied in obvious ways to obtain the desired loading of the capacitor C_X during the propagation delay. Any kind of controllable semiconductor switch or other controllable switches may be used. The same applies to the other switching transistors MP1 and MN1. The trigger circuit 30 in FIG. 5 may be replaced by digital delay circuit based on a counter circuit, timer circuits, flip flop circuits and/or monostable circuits. The same applies to the delay circuit 10, provided that the alternative digital circuit also provides a propagation delay depending on the supply voltage V_DD. Accordingly the preferred embodiment may vary within the scope of the attached claims.