With microprocessor clock speeds rising above 1 GHz, the CPU's heatsink comes under increasing scrutiny as a source of electromagnetic interference (EMI). Often, this problem isn't addressed until after the microprocessor has been designed in and the product is undergoing EMI tests. If EMI problems are discovered at that point, they're addressed with Band-Aid solutions, such as adding shielding hardware. Naturally, such fixes slow product development and add cost. Moreover, they may be ineffective in future product spins with newer processors.
Responding to this dilemma, re-searchers at IBM in Research Triangle Park, N.C., have devised a novel microprocessor socket that separates heatsink grounding design from processor clock speed. In effect, the grounded socket can eliminate the need to change grounding schemes when a new processor is designed into the next generation of a product. This work is described in a paper titled, "Improved Grounding Method for Heatsinks of High Speed Processors," presented at the Electronic Components and Technology Conference held May 29 through June 1 in Orlando, Fla.
By incorporating grounding contacts into the processor's socket, Joseph Diepenbrock and his colleagues at IBM were able to reduce the inductance from heatsink to ground, significantly lowering the RF voltages present on the heatsink (Fig. 1). In addition to decoupling the grounding design from clock speed, the new heatsink grounding technique makes grounding independent of the mechanical assembly used to secure the heatsink to the pc board.
The prototype socket, de-signed and built to IBM's conceptual specification by Foxconn International Inc. of Sunnyvale, Calif., is slightly larger than its ungrounded equivalent. But the increase in socket area is just 16%.
In developing their socket, the IBM researchers came upon some surprising results. For one, they found that the main source of EMI in their test system was not the processor's clock but an external bus carrying a much lower frequency signal.
Testing a variation of their basic socket design led to an interesting conclusion about the relative importance of ground-path inductance versus shielding aperture size. Both socket designs employ a series of grounding contacts embedded in the periphery of the socket. Between each of the contacts, there is a gap that represents an aperture through which RF energy may be expected to leak. The aperture is the same for each of the socket designs.
There's just one difference between the two sockets. The first version employs contacts with sharp "nubs" to break through oxides that accumulate on the heatsink, while the second employs flat contacts.
Researchers used the two grounded sockets as well as an ungrounded socket in experiments where they measured heatsink impedance to ground over the 50- to 5000-MHz range. They discovered that both of the grounded socket designs produced much lower impedance readings than the ungrounded heatsink (Fig. 2).
However, they also discovered some differences in the performance of the two grounded sockets. The one with flat contacts exhibited lower impedance at the lower test frequencies while producing some impedance peaks at the higher frequencies. According to Diepenbrock, the difference in impedance measurements stems from the change in inductance that results from changing from nubbed to flat contacts.
Because aperture size was the same in both grounded sockets, Diepenbrock asserts, the inductance of the heatsink's ground connection determines the level of RF voltage on the heatsink and the associated radiation. That's in contrast with previous findings by other researchers, who have held that shielding aperture size dictates the degree of heatsink radiation.
For a copy of the IBM paper, contact IEEE's order department at (800) 678-4333, or see www.cpmt.org/proceedings. Or to contact the authors directly, contact Joseph C. Diepenbrock at (919) 543-8804 or [email protected].