Remix.run Logo
▲ IAmNotACellist an hour ago

I'd be fascinated to know the exact nature of the failure. What analog processes cause a CMOS memory IC that we like to think of as being discrete to be consistently wrong at reading its stored values? I'd like to know all the way down to the interaction between the gates and traces. Did some pathway erode due to constant interchange of charge carriers? Did thermal cycles or stress cause one of the mechanical interconnects on the package to fracture and no longer conduct? If so, which one?

According to AI, one type of failure mode in those early aerospace CMOS chips was sudden parasitic current in the PNPN junction being caused by a high energy particle, shorting paths to ground melting metal. Though that sounds rare. Another option is a total ionizing dose of radiation over time, which creates electron-hole pairs in the gate oxide, leading to trapped positive charge. That makes NMOS get closer and closer to turning on when they shouldn't, and PMOS requiring more voltage to turn on. It's fascinating to me that there came a specific moment in time when the chip managed to work, and the next clock cycle it didn't. I'd love to see the I_D vs. V_GS charts showing a before and after. So I had Claude make a plausible approximation. You can too by asking your AI agent of choice by basically just pasting in my comment!

▲raj-shekhar 34 minutes ago | parent | next [-]

I tried to find the exact root cause. NASA said they could not determine the cause of the chip failure with certainty, so we can only speculate.

This is what I found as part of my research:

NASA: "Engineers can't determine with certainty what caused the issue. Two possibilities are that the chip could have been hit by an energetic particle from space or that it simply may have worn out after 46 years." https://science.nasa.gov/blogs/voyager/2024/04/04/engineers-...

IEEE Spectrum, citing project scientist Linda Spilker, reported that the memory readout showed "a stuck bit, fixed at the same binary value" in that 256-bit chunk: https://spectrum.ieee.org/voyager-1

EE Journal reported JPL's chief engineer naming the part as an RCA CD4061A, a 256 x 1-bit CMOS SRAM. Losing it took out 256 of the data computer's 8,192 words, which matches NASA's 'about 3%': https://www.eejournal.com/article/jpl-software-update-rescue...

▲alhadrad an hour ago | parent | prev [-]

It is most certainty cumulative radiation effects degrading the gates as you have eluded. This problem still exists for on orbit power electronics.