|
|
|
|
|
|
Answers Database
Virtex: Bypass Capacitors -- what are the requirements?
Record #777
Product Family: Hardware Virtex: Bypass Capacitors -- what are the requirements? Problem Description: Urgency: Standard General Description: Recommendations for decoupling Xilinx devices. Resolution 1: Theory Resolution 2: Virtex Specific Solution 1: The standard Vcc-decoupling rule for modern CMOS -ICs is to mount one low-inductance decoupling capacitor of 0.01 to 0.1 uF very close to each Vcc pin. The purpose of these decoupling capacitors is to provide the fast changing dynamic supply current, especially when capacitively loaded outputs are switching. The big power-supply capacitor, and even the 100uF board decoupling capacitor cannot perform this function for two reasons: Those big capacitors have an unavoidable internal series-inductance which makes them incapable of supplying fast-changing current pulses with nanosecond rise and fall-times. These capacitors are also physically too far away from the devices that need access to an instant current reservoir. Even a good power/ground-distribution network has unacceptable inductance over distances of several inches. As CMOS devices have become faster and as the number of outputs increases, good decoupling is becoming more important. Current pulses with 1 ns rise or fall-time must be treated like GHz signals, and the digital pc-board designer might learn something from looking inside a UHF-TV tuner. Here are some quantitative data: Let us assume a device with 40 outputs switching simultaneously (or within a few nanoseconds of each other ), with each output driving a 100 pF load. With an output swing of 4 V and all outputs switching in the same direction, each such transition consumes 0.016 uCoulomb. If four decoupling capacitors of 0.01 uF have to supply this charge, they will drop their voltage by 0.016/0.04 = 0.4 V. That is barely acceptable, and four 0.1 uF decoupling capacitors would be preferred, since they reduce this drop by an order of magnitude. Capacitors larger than 0.1 uF tend to have more internal series-inductance, so they are actually inferior for this application. Another way to look at dynamic current requirements is to start with power dissipation: A 3-W device clocked at 40 MHz does not consume a steady 600 mA, but might consume 3 A for 5 ns, and very little for the rest of the time. As explained above, these 0.015 uCoulomb require at least four 0.1 uF decoupling capacitors, and 0.1 uF would be better. The need for Vcc decoupling varies with the amount of internal logic, and the number of outputs switching ( worst-case ) simultaneously, and their capacitive load. The classical rule of one low-inductance 0.01 to 0.1 uF at each Vcc pin is still a good guideline. Solution 2: For bypassing high-speed, high-density designs (Virtex included), see (Xilinx XAPP158, page 2) The app note is intended as a guideline for bypassing. Designers often wonder why we are suggesting such large caps (47uF, 470uF), and the answer is relatively simple: the series R and L of smaller value capacitors (4.7uF) is so high to make them useless (better off removing them altogether). The other factor here is that we are trying to generalize the millions of possible designs on the part. What bypassing scheme is the safest to recommend if you have no idea how the part is to be used? To do it absolutely by the book, the bypassing for the core, and the io depends on the system clock, the data rate, the io standard chosen, the printed circuit board signal integrity design rules, the environmental regulations (FCC Part A or Part B? VDE?, CSA? BellCore?). In order to come up with more specific information about bypassing, the designer needs to know the answer to the above questions: system clock rate(s), system io data rate(s), io standard(s) used, intended application, regulatory constraints (Bellcore? ITU-T? ITU-R? DataComm? Commercial?), environmental temperature operating range (-40C to +75C?, 0C to 55C?). Given this information, one can make an appropriate modification to what appears in app note 158. End of Record #777 - Last Modified: 01/27/00 12:34 |
| For the latest news, design tips, and patch information on the Xilinx design environment, check out the Technical Tips! |