Soft Errors ... They're Baaaaack!

When it comes to device reliability, the semiconductor industry is largely focused on "hard" errors, where a failure results from some physical degradation of the device. Common reliability problems include electromigration and stress migration in metal interconnects, hot carrier injection at the gate, antenna charging, and damage from electrostatic discharge. These were the main focus of the recent International Reliability Physics Symposium (IRPS).

What may be surprising to some is a new focus on soft error rates (SERs). Soft errors were big news in the late '70s and early '80s, when it became known that alpha particles generated by tiny amounts of naturally occurring radioactive impurities found in many packaging materials were causing errors in DRAMs. Alpha particles don't do any permanent circuit damage, but they do upset internal data states in the device, which shows up as an error. That problem was addressed by improving the purity of molding epoxies, solders and other materials.

As devices have shrunk and operating voltages have dropped, however, soft errors have again reared their ugly head. In addition to alpha particles emitted by radioactive impurities — uranium and thorium mostly — soft errors are now known to be caused by cosmic rays from outer space, and from the results of collisions between cosmic rays and boron. The impact of these soft errors is huge — much larger than those associated with "hard" errors, said Robert Baumann, reliability scientist and distinguished member of the technical staff at Texas Instruments (Dallas). Baumann, who also heads International SEMATECH 's SER working group, led a day-long session on SERs and a three-hour panel discussion at IRPS.

When 10B, a particular species of boron commonly found in BPSG, gets hit by a cosmic neutron, the nucleus splits apart violently, generating an alpha particle and a lithium particle. Those particles can generate soft errors. (Source: TI)

Baumann explained that, for most applications, that's a completely acceptable failure rate in that 100,000 FITs is roughly equivalent to one failure per year (assuming 24-hour-per-day operation). In a mobile phone application, for example, where there is one chip in the system, that might mean you'd lose one call over the course of a year ... not too bad.

However, for other applications such as network servers, it's unacceptable. "There, the problem is they take this 100,000 FIT part and they put 20 of them on a board, and 20 or 30 boards in a system," Baumann said. "Now you've got a system that fails every day. These are the people that are really, really up in arms about SER. We've done a lot of work to address those needs."

Part of that work has been to demand higher-purity processing and packaging materials, and to develop a metrology system that can measure them. "The level to which we have to reduce alpha emissions is two to three orders of magnitude lower than what it used to be," Baumann noted. "That's really the fundamental materials challenge."

To meet that challenge, it's necessary to be able to measure an ultralow alpha flux. So far, that ability doesn't exist. To address that, Baumann and associates authored a SEMATECH technology transfer document, released last year, titled "Call for Improved Ultra-Low Background Alpha-Particle Emission Metrology for the Semiconductor Industry."

"In this document, we asked that third-party vendors give us the capability to measure 0.0001 alphas per cm² per hour," Baumann said. "What that says is from one square meter of material, we want to be able to catch every alpha coming off that down to one alpha per hour. That's at least an order of magnitude lower than the industry has ever asked for before. That's a significant challenge."

Baumann also was instrumental in generating a JEDEC standard, JESD89, titled "Measurement and Reporting of Alpha Particles and Terrestrial Cosmic Ray-Induced Soft Errors in Semiconductor Devices," which is a set of standardized methods of testing and reporting data. "It explains you cannot report SER data unless you have a number for cosmic rays, a number for boron 10 activations, and a number for alpha particles," Baumann said. "It basically puts everybody on equal footing in terms of saying, Hey, if you're going to report soft error rates, you've got to do it right."

It will never be possible to stop cosmic rays from outer space (unless you live in a deep cave where they are blocked). "These are basically energetic particles from space — mostly protons," he said. "When they go through the atmosphere, they get filtered. By the time they hit our level, they're mostly neutrons."

What's interesting is what happens when these neutrons hit the device — particularly boron 10 (¹⁰B). "Because I was doing some other things actually on alpha particles, I kind of stumbled across the fact that boron 10 has a huge cross section. It's well known in the nuclear industry, but the semiconductor industry hadn't really latched onto that. Unlike most other isotopes, when boron 10 captures a neutron, the nucleus splits apart violently. The particles that are emitted when it splits apart are basically an alpha particle and a lithium particle, both of which can cause a soft error," Baumann said. The Figure shows the result of a ¹⁰B activation by a low-energy neutron.

This discovery, first published in 1995, led TI to eliminate boron in the BPSG (borophosphosilicon glass) layer, a conformal dielectric film typically used on top of the transistors before the first level of metal is deposited. "Since the BPSG was in close proximity to the device silicon, when these little nuclear explosions occurred, they induced soft errors. This was a completely new mechanism and it turned out, in the 0.18 technology at least, to be a dominant component," Baumann said. By removing the boron in BPSG, a 10× reduction in soft errors was seen. Boron used elsewhere in the device, such as for doping, is not of concern because it is in far lower concentrations than in BPSG, or of a different isotope of boron that does not cause the damaging radiations.