TECHNICAL INFORMATION DECOUPLING: BASICS by Arch Martin AVX Corporation Myrtle Beach, S.C. Abstract: This paper discusses the characteristics of multilayer ceramic capacitors in decoupling applications and compares their performance with other types of decoupling capacitors. A special high-frequency test circuit is described and the results obtained using various types of capacitors are shown. Introduction LC CC ZC = Î RC2 + (XC - XL)2 RC XC = 1 2pfCC XL = 2pfLC K 3 1 Z2 ZX Z Fig. 2. Total capacitor impedance Z The rapid changes occurring in the semiconductor industry are requiring new performance criteria of their supporting components. One of these components is the decoupling capacitor used in almost every circuit design. As the integrated circuits have become faster and more dense, the application design considerations have created a need to redefine the capacitor parameters and its performance in high-speed environments. Faster edge rates, larger currents, denser boards and spiraling costs have all served to focus upon the need for better and more efficient decoupling techniques. As integrated circuits have grown, so has the demand for multilayer ceramic capacitors. z BRIDGE EXCITATION VARIABLE E & F UNKNOWN The phenomenal growth of multilayer ceramic capacitors over the last few years has been a result of their ability to satisfy these new requirements. We at AVX are continually studying these new requirements from the application view in order to better define what is required of the capacitor now and in the future, so that we can develop even better capacitor designs. Some results of these studies are the subject of this paper. Background A capacitor is an electrical device consisting of two metal conductors isolated by a nonconducting material capable of storing electrical charge for release at a controlled rate and at a specified time. Its usefulness is determined by its ability to store electrical energy. An equivalent circuit for capacitors is shown in Fig. 2. This equivalent circuit of three series impedances can be represented by one lumped impedance which is used as a measure of the capacitance (Fig. 3). In other words, the amount of coulombs stored are not measured; what is measured is the lumped impedance and from this value an equivalent capacitance value is calculated. ZS ZX = ZX = Z1Z3 Z2 (When ZX ,,,Z3) uZ1u . uZ3u Z2 UX = U1 + U3 - U2 Fig. 3. Basic impedance bridge Thus capacitance as measured is actually a combination of the capacitive reactance, the inductive reactance and the equivalent series resistance. As shown in Fig. 4, all three of these series impedances vary differently with frequency. Since all three vary at different rates with frequency, the capacitance calculated from the resultant impedance is made up of different components at different frequencies. This can be seen by increasing the lead length of a capacitor while it is being measured at 1 MHz on an equivalent series capacitance bridge and watching the capacitance increase. Increased inductance (increased lead length) actually increases the capacitance value read by the capacitance bridge (Fig. 5). IMPEDANCE Z XL ESR XC FREQUENCY Fig. 4. Variation of impedance with frequency L 1/vC jVL XC J XCeq = XC + XL 1 = jvCeq XCeq 1 Ceq = 1 Ceq = Ceq = XC 1 + jvL jvC 1 C - v2L IMPEDANCE, V XL 1.0 C XL 0.1 1 - v2LC C C 1 - v2LC AS "L" INCREASES "Ceq" INCREASES Fig. 5. Effects of inductance 0.02 RC 1 10 100 MEGAHERTZ Fig. 6. Major impedance components The major components of a typical impedance curve vs. frequency are shown in Fig. 6. Below resonance, the major component is capacitive reactance, at resonance it is equivalent series resistance, and above resonance it is inductive reactance. In decoupling today’s high-speed digital circuits the capacitor is primarily being used to eliminate high-speed tran- sient noise which is above its resonance point, an area where inductive reactance is the major impedance component. In these applications it is desirable to maintain as low an inductance or totol impedance as possible. For effective and economical designs it is important to define the performance of the capacitor under the circuit conditions in which it will be used. High-Frequency Testing The test schematic shown in Fig. 7 can be used to determine the performance of various styles of capacitors at high frequencies. Lead lengths are minimized through the use of the test fixture shown in Fig. 8. This test set-up is intended to duplicate the performance of the capacitor under actual use conditions. The Hewlett-Packard current generator supplies high-frequency pulses equivalent to an actual digital IC circuit. Edge rates of 200 mA/10 ns or 200 mA/5 ns are fed to the capacitor to determine its performance. Edge rates in the 5 to 10-ns range were chosen for testing since transients in this range are common to many designs such as shown in Fig. 9, where transients in the 5 to 10-ns range occur regularly on a DEC VAX 11/780 memory board using Mostek 4116 dynamic RAMs. 100 90 10 0% SCALE: HORIZONTAL = 1mS/DIV VERTICAL = 5MV/DIV 50V TERM TEKTRONIX 001-0049-01 50V HP8012B TEKTRONIX 475A OR EQUIVALENT INDUCTANCE TEST FIXTURE Fig. 7. High-frequency test schematic X X COPPER-CLAD PC BOARD. X INDICATES TEST AREAS Fig. 9. Transients from VDD to VSS in 256K byte dynamic memory (Mostek 4116) in DEC VAX 11/780 All three components (capacitive, resistive, and inductive) are evident on the scope trace (Fig. 10) obtained by using the test circuit of Fig. 7. Thus the performance of a capacitor can be described under typical use conditions for capacitance (C = Idt⁄di), equivalent series resistance (ESR = VA/I), and inductance (L = VL dt⁄di). This information can then be used to select a capacitor having the required characteristics for the application. 100 90 dv __ (C = 1dt ___ ) dv dt VL dt (L=VL __ ) di 10 0% __A ) VR(ESR = V 1 SCALE: HORIZONTAL = 50NS/DIV VERTICAL = 50MV/DIV COAXIAL CABLE TO SCOPE AND PULSE GENERATOR Fig. 8. Inductance test fixture Fig. 10. Scope trace of 0.22-µF film capacitor using test setup of Fig. 7 and 200 mA/10 ns input 100 90 100 90 10 0% 10 0% SCALE: HORIZONTAL = 50NS/DIV VERTICAL = 100MV/DIV Fig. 11 (A). Scope trace of 0.1-µF tantalum capacitor using test setup of Fig. 7 and 200 mA/10 ns input SCALE: HORIZONTAL = 50NS/DIV VERTICAL = 100MV/DIV Fig. 11 (B). Scope trace of 0.1-µF multilayer ceramic capacitor using same test setup and input Test Results Figures 11(A) and 11(B) compare the results of testing a 0.1-µF rated multilayer ceramic capacitor with a 0.1-µF rated tantalum capacitor. The slope of the curve, dv/dt, is a measure of the capacitance in the time domain of interest. For the ceramic this is approximately 0.1 µF but for the tantalum the value is 0.05 µF or approximately half the capacitance of the ceramic. Another way to state this is at constant current: C dv C dv Tantalum dt = I = Ceramic dt C 0.2 = C 0.1 T 50 x 10-9 C 50 x 10-9 C = T 1 ⁄2 C C The induced voltage generated in response to the 200 mA/10 ns edge rate can be used to determine the inductance of the capacitor. Under these conditions the tantalum shows 2.5 times the inductance of the multilayer ceramic (MLC) capacitor: L = V dt L di Ceramic Tantalum -9 L = 0.1 10 x 10 0.2 L = 5.0 nH -9 L = 0.25 10 x 10 0.2 L = 12.5 nH The ESR is obtained from the minimum voltage reached after the voltage spike. In the case of the ceramic the ESR is too small to measure on the scale used, i.e., less than 5 mΩ. The ESR for the tantalum can be determined with the VR = 150-mV reading giving an ESR = 750 mΩ. Both the MLC and the tantalum capacitors were tested with equal lead lengths. Removal of the leads reduces the inductance obtained for the MLC (now an MLC chip capacitor) to VL = 20 mV for an inductance of 1 nH. This test procedure can be used to compare the performance of various styles of capacitors. Fig. 12 compares the performance of MLC and a film capacitor. Both units show dv/dt equivalent to the capacitance values read at 1 kHz (0.22 µF for the film and 0.20 µF for the MLC). The MLC shows a voltage spike of 75 mV vs. the 275-mV spike for the film, i.e., the film had 3.7 times the inductance of the MLC. In addition, the MLC had half the ESR of the film capacitor. The initial voltage spikes found from an input of 200 mA/5 ns edge rate are given in Table I for various capacitor values and styles. The equivalent inductances for these values are shown in Table II. These results are based on essentially zero lead lengths for all types. From a practical standpoint, the lead length inductances shown in Table III must be added to values of Table II in order to mate the capacitor with the PC board. CAPACITOR FILMS: BRAND A BRAND B BRAND C BRAND D 100 90 0.22mF FILM 0.20mF CERAMIC 10 0% SCALE: HORIZONTAL = 20NS/DIV VERTICAL = 100MV/DIV Fig. 12. Comparison of 0.2-µF MLC with 0.22-µF film capacitor CAPACITOR FILMS: BRAND A BRAND B BRAND C BRAND D 170 110 180 300 240 120 300 — — 120 — 230 470 140 350 — — — — — TANTALUMS: BRAND E — 380 — 380 — ALUMINUMS: BRAND F — 630 — 400 — — 70 — 70 90 — — — 50 — 40 50 — — — 80 130 140 210 — AVX MLC’S CONFORMAL RADIALS MOLDED RADIALS DIP’S MOLDED AXIALS Table I. Voltage spikes (mV) from 200 mA/5 ns edge rate with essentially zero lead lengths .01µF .1µF .22µF 1.0µF 2.2µF 4.3 2.8 4.5 7.5 6.0 3.0 7.5 — — 3.0 — 5.8 11.8 3.5 8.8 — — — — — TANTALUMS: BRAND E — 9.5 — 9.5 — ALUMINUMS: BRAND F — 15.8 — 10.0 — AVX MLC’S CONFORMAL RADIALS MOLDED RADIALS DIP’S MOLDED AXIALS .01µF .1µF .22µF 1.0µF 2.2µF — 1.8 — 1.8 2.3 — — — 1.3 — 1.0 1.3 — — — 2.0 3.3 3.5 5.3 — Table II. Inductances (nH) from 200 mA/5 ns edge rate with essentially zero lead lengths TYPICAL LEAD LEAD LEAD LEAD VL IMP. CAPACITOR TYPE LENGTH L (MM) (NH) (MV) (Ω) FILMS: BRAND A BRAND B BRAND C BRAND D 3.81 2.54 3.81 2.54 1.5 1.0 1.5 1.0 60 40 60 40 0.9 0.6 0.9 0.6 TANTALUMS: BRAND E 5.08 2.0 80 1.3 ALUMINUMS: BRAND F 4.45 1.8 70 1.1 2.54 1.0 40 0.6 2.54 5.08 1.0 2.0 40 80 0.6 1.3 7.62 3.0 120 1.9 AVX MLC’S CONFORMAL RADIALS MOLDED RADIALS DIP’S MOLDED AXIALS Table III. Typical lead lengths required to connect capacitor to PC board Decoupling The above capacitor models can be used to optimize the decoupling of integrated-circuit designs. As an example, the decoupling requirements of dynamic RAMs will be discussed. Dynamic RAMs have large transients which are generated during their refresh cycle. These large transients require careful attention to the decoupling techniques to avoid “V bump” or “soft” error problems. POWER-SUPPLY DRIFT + NOM. VOLTAGE BULK CURRENTS HIGH FREQ. NOISE di V = L __ dt dv I = C __ dt Fig. 13. Three factors causing voltage variations The function of the capacitor can be illustrated by referring to Fig. 13. Through time, the three factors causing voltage variations are: dc drift, bulk variations, and switching transients. Dc drift is independent upon power-supply design and not on the board level decoupling. Bulk variations come about by the current demands of recharging the internal storage cells during the refresh cycle. Transient “noise” comes from switching currents internal to the IC chip. The total of these three voltage variations must be maintained within the allowed tolerance for the IC device. In other words, if the voltage drops below the operating margin of the IC, a “soft” error will occur. The board low-frequency surge current to be supported by the bulk capacitance is effectively the IC’s average refresh active current for the length of a refresh cycle. This can be determined by referring to the IC spec (and assuming worst-case that stand-by current is zero and active current is maximum). The traditional approach for calculating bulk capacitance requirements is to multiply the capacitance needed per package times the number of packages on the board. This arrives at the total capacitance required, which is then approximated with large-value capacitors around the periphery of the circuits. It is more effective to distribute this capacitance throughout the design with smaller value capacitors whose combined total meets or exceeds the bulkcurrent requirement. Distribution of the bulk requirement to capacitors adjacent to the current requirements (the IC chip) is beneficial in reducing inductance and resistive voltage drops. Slight increases in the distributed transient capacitor values can typically eliminate the need for large capacitors to supply bulk current. Transient noise is commonly called in the industry “V bump,” and reflects the supply transient induced by the chip itself on the decoupling capacitor when various clocks fire on-chip and drive on-chip capacitance associated with that event (such as address decoding). Since actual loads switched are small (20 pF), the size of the decoupling capacitor is less important than its inductance. Using capacitors for reducing the line “noise” that comes from switching internal to the IC chip requires low inherent inductance within the decoupling capacitor and effective board design. Multilayer ceramic capacitors are available in values high enough to meet distributed bulk requirements while maintaining low inductance at high frequencies. When a capacitor is mounted on a board, lead lengths and board lines (device to capacitor to ground) are a major source of inductance. This inductance must be minimized to obtain good decoupling performance under high-speed transient conditions. Minimum lead lengths, wiring, and gridding of power supplies and ground with alternate parallel paths are important as is the quality of the capacitor (Fig. 14). The use of multiple capacitors instead of a few large bulk capacitors can be used to decrease line lengths and to increase path numbers (gridding) for reduced inductance and more effective surge-current availability. 0.9" 0.2" 0.1mF (0.2") Fig. 14. Circuit board line and capacitor lead lengths USA EUROPE ASIA-PACIFIC AVX Myrtle Beach, SC Corporate Offices AVX Limited, England European Headquarters AVX/Kyocera, Singapore Asia-Pacific Headquarters Tel: 843-448-9411 FAX: 843-626-5292 Tel: ++44 (0) 1252 770000 FAX: ++44 (0) 1252 770001 Tel: (65) 258-2833 FAX: (65) 350-4880 AVX Northwest, WA AVX S.A., France AVX/Kyocera, Hong Kong Tel: 360-699-8746 FAX: 360-699-8751 Tel: ++33 (1) 69.18.46.00 FAX: ++33 (1) 69.28.73.87 Tel: (852) 2-363-3303 FAX: (852) 2-765-8185 AVX North Central, IN AVX GmbH, Germany - AVX AVX/Kyocera, Korea Tel: 317-848-7153 FAX: 317-844-9314 Tel: ++49 (0) 8131 9004-0 FAX: ++49 (0) 8131 9004-44 Tel: (82) 2-785-6504 FAX: (82) 2-784-5411 AVX Mid/Pacific, MN AVX GmbH, Germany - Elco AVX/Kyocera, Taiwan Tel: 952-974-9155 FAX: 952-974-9179 Tel: ++49 (0) 2741 2990 FAX: ++49 (0) 2741 299133 Tel: (886) 2-2696-4636 FAX: (886) 2-2696-4237 AVX Southwest, AZ AVX srl, Italy AVX/Kyocera, China Tel: 480-539-1496 FAX: 480-539-1501 Tel: ++390 (0)2 614571 FAX: ++390 (0)2 614 2576 Tel: (86) 21-6249-0314-16 FAX: (86) 21-6249-0313 AVX South Central, TX AVX Czech Republic, s.r.o. AVX/Kyocera, Malaysia Tel: 972-669-1223 FAX: 972-669-2090 Tel: ++420 (0)467 558340 FAX: ++420 (0)467 558345 Tel: (60) 4-228-1190 FAX: (60) 4-228-1196 Elco, Japan AVX Southeast, NC Tel: 045-943-2906/7 FAX: 045-943-2910 Tel: 919-878-6223 FAX: 919-878-6462 Kyocera, Japan - AVX AVX Canada Tel: (81) 75-604-3426 FAX: (81) 75-604-3425 Tel: 905-564-8959 FAX: 905-564-9728 Kyocera, Japan - KDP Tel: (81) 75-604-3424 FAX: (81) 75-604-3425 Contact: NOTICE: Specifications are subject to change without notice. Contact your nearest AVX Sales Office for the latest specifications. All statements, information and data given herein are believed to be accurate and reliable, but are presented without guarantee, warranty, or responsibility of any kind, expressed or implied. Statements or suggestions concerning possible use of our products are made without representation or warranty that any such use is free of patent infringement and are not recommendations to infringe any patent. The user should not assume that all safety measures are indicated or that other measures may not be required. Specifications are typical and may not apply to all applications. © AVX Corporation A KYOCERA GROUP COMPANY http://www.avxcorp.com S-DB2.5M1194-C