WAKEFIELD, Mass. — Supporters of the CompactPCI bus call some recent hot-swap hiccups rare and easily remedied, but at least one critic says they could undermine confidence in the bus for high-availability applications.
The PCI Industrial Computer Manufacturers Group (PICMG), based here, moved earlier this week to revise the CompactPCI Hot Swap spec by forming a technical committee to look at 3.3-volt operation and the "challenges with pin bouncing." Then Motorola Computer Group (MCG; Tempe, Ariz.) detailed those challenges for PICMG members.
Sizing up the problem
A critic of CompactPCI, who spoke on the condition of anonymity, said that a hot-swap problem could make some CompactPCI users reevaluate their choice. "Hot-swap is the whole hook they've hung their hat on" for high-availability systems, he said. But PICMG president Joe Pavlat dismissed such concerns as "fear mongering" by CompactPCI's critics, and said the problem is "not a mountain. It's a nanohill."
Meanwhile, AMP Inc./Tyco Electronics (Harrisburg, Pa.) has pulled a patent out of its pocket and says it can cure those pin bounce blues. The company is expected to announce its solution — a line of connectors called QuietMate — early next month. Company officials declined to discuss the connectors, but they are expected to bring samples to simultaneous shows early next month in California: DesignCon, in San Jose, Calif., and Bus and Board, in San Diego.
Connector pin bounce has been causing transaction errors at the site of a Motorola Computer Group customer, a large maker of telecom equipment, when CompactPCI boards are inserted. "It causes the bridge chips to freak out," the CompactPCI critic said. Contact oscillation at the molecular level is mistaken for a data transmission, the source said, and this generates a parity error.
But only one PCI bridge chip has displayed the error, said the source, who refused to identify the IC. Other PCI bridge chips "do not exhibit the failure at all," he said, calling the chip at issue "a known problem child."
PICMG's Pavlat, who is also director of strategic planning at Motorola Computer Group's Monterey Design Center, downplayed the problem's impact on the bus, saying there is "no cause for alarm."
"MCG has verified these rare failures in the lab and with the help of engineers at AMP Inc. has determined their cause," he said. "We call it nanobounce, and it appears to be a nanoproblem — and one that occurs only once in a proverbial blue moon."
Pavlat described the circumstances surrounding the telecom manufacturer's problem as unusual. "Our customer was transmitting a large amount of ATM data on a very heavily loaded system. He was hammering the bus at in excess of 95 percent of its theoretical capacity and he documented — and we've confirmed — that one bus transaction in 1,000 will fail and generate a parity error.
"In a more typically loaded system, maybe it could happen as infrequently as once in a million times," he said, "but the connector is only rated for a couple of hundred insertions, which is more than the expected number of hot-swap events that would occur in a typical system over its lifetime. So, at this point, it's more an issue of scientific and engineering curiosity than a cause for alarm."
Pin bounce itself remains a mystery. "The underlying cause is not clearly understood," Pavlat said. "We're talking about a signal bounce on the order of 5 to 10 nanoseconds.
"From a physics standpoint, it's an interesting and heretofore unseen phenomenon. It appears to be caused by microscopic surface irregularities of either the connector pin or the socket," Pavlat said.
Meanwhile, the PICMG will continue to examine the bridge-chip issue, Pavlat said. "It's true we see it with some manufacturers' chips and not with others, but we haven't taken enough data yet," he said. "As we do a more detailed study, we'll do our best to make this data known."
Pavlat said AMP may already have a solution, even before the hot-swap CompactPCI community addresses the problem. "They've invented a secret sauce, a resistive coating for the end of the pins that we have tested," he said. "And as far as we can tell, it eliminates the problem altogether, but we do not believe it is the only way to address this transaction error. Other solutions, including careful software control during hot swap, may prove effective."
Since the AMP connector is patented, he said, "it is unlikely it will be written into any future version of the hot-swap spec, because we want to promote open technology."
