When I was designing ethernet cards for 3Com, my department wouldn't use electrolytic capacitors. Not aluminum, not tantalum. Tantalum performs better, but there were really bad environmental and human rights problems with the way they mine it, and eventually there were shortages. We used multilayer ceramics (MLC), surface mounted, for everything. You got as much supply bypass effect from a 2.2 uF MLC as from a 22 uF wet tantalum or any size wet aluminum. MLCs have the same wear-out mechanism as ICs, electromigration. That's why you need to pay attention to the current rating when you design them in. But it's nothing like the way electrolytics outgas and dry out.
The interesting thing is really that modern CMOS circuitry (that's practically everything electronic) consumes its power in very short bursts. You're delivering a "DC" supply but the actual power consumption is a broadband phenomenon with much of its energy in the 100 MHz to GHz range. That's why there are bypass caps on the board at all. The voltage regulator in the power supply responds in milliseconds and the bypass caps have to fill in the microsecond and nanosecond sags. At those frequencies, the ESR of an aluminum electrolytic cap is so high it might as well not be there.
(That's why, by the way, there is no point in putting a car stereo stiffening capacitor in a computer case. Those half farad oil cans with their hundred nanohenry leads aren't fast enough to do much but keep your power light on for a while. But they do improve the damping factor in the bass end of a fancy car stereo, by mitigating the series resistance of the car battery and fuses. Mount them next to the amp and use thick copper ribbon, not wire.)
Later, I did a big PowerPC CPU board for a router company. The most important bypass capacitor on that board was the board itself. Many square inches of solid (except vias of course) power and ground planes 1.7 mils apart. That's got the lowest ESR of any capacitor there is. By then there were SMT MLCs in the 200 uF range. I believe small camcorders forced their development. When the PPC750L jumps from sleeping to running, it draws a current spike about 3 ns wide and about 85 Amperes high. I would assume pentium 4 and athlon 64 do the same kind of thing. With sub 1 uF MLCs and all the aluminum electrolytics you can fit anywhere near it, that 2V plane sags to less than 1.2V when that spike goes by, and the CPU crashes. And there's no difference if you remove all the electrolytics.
In our solution, we used those multilead MLCs that you see soldered to the top of some CPU packages, the planes in the board, an assortment of small MLCs, and 100 uF MLCs near the four corners of the CPU package. The multilead MLCs are really cool: if you simulate them in a 3d field solver you can see how the inductance of each lead partially cancels out the inductance of its neighbors. Hardest part is making room for enough vias not to throw that effect away when you connect them to buried planes.
So I wonder if anybody has tried recapping these motherboards with MLCs and copper foil, and screw the aluminum cans.
The interesting thing is really that modern CMOS circuitry (that's practically everything electronic) consumes its power in very short bursts. You're delivering a "DC" supply but the actual power consumption is a broadband phenomenon with much of its energy in the 100 MHz to GHz range. That's why there are bypass caps on the board at all. The voltage regulator in the power supply responds in milliseconds and the bypass caps have to fill in the microsecond and nanosecond sags. At those frequencies, the ESR of an aluminum electrolytic cap is so high it might as well not be there.
(That's why, by the way, there is no point in putting a car stereo stiffening capacitor in a computer case. Those half farad oil cans with their hundred nanohenry leads aren't fast enough to do much but keep your power light on for a while. But they do improve the damping factor in the bass end of a fancy car stereo, by mitigating the series resistance of the car battery and fuses. Mount them next to the amp and use thick copper ribbon, not wire.)
Later, I did a big PowerPC CPU board for a router company. The most important bypass capacitor on that board was the board itself. Many square inches of solid (except vias of course) power and ground planes 1.7 mils apart. That's got the lowest ESR of any capacitor there is. By then there were SMT MLCs in the 200 uF range. I believe small camcorders forced their development. When the PPC750L jumps from sleeping to running, it draws a current spike about 3 ns wide and about 85 Amperes high. I would assume pentium 4 and athlon 64 do the same kind of thing. With sub 1 uF MLCs and all the aluminum electrolytics you can fit anywhere near it, that 2V plane sags to less than 1.2V when that spike goes by, and the CPU crashes. And there's no difference if you remove all the electrolytics.
In our solution, we used those multilead MLCs that you see soldered to the top of some CPU packages, the planes in the board, an assortment of small MLCs, and 100 uF MLCs near the four corners of the CPU package. The multilead MLCs are really cool: if you simulate them in a 3d field solver you can see how the inductance of each lead partially cancels out the inductance of its neighbors. Hardest part is making room for enough vias not to throw that effect away when you connect them to buried planes.
So I wonder if anybody has tried recapping these motherboards with MLCs and copper foil, and screw the aluminum cans.
Comment