A helium nuclei, which contains two protons and two neutrons. Alpha
particles are emitted from larger atoms as a result of radioactive decay.
An alpha particle will only travel a few centimeters in air, or about 25
microns in silicon, before interacting with the matter it is travelling
through.
Bathtub Curve
The bathtub curve refers to the failure rate of many manufactured
devices when viewed over the lifetime of the device. Many devices fail
within a short period of time due to manufacturing
defects. This failure rate decreases with time until a relatively constant
failure rate is achieved. This constant failure rate applies to the normal
working life of the device, after which time the failure rate starts to
increase to due parts wearing out.
Clean room
Semiconductors are constructed within a cleanroom, which is a very
precisely controlled (and clean) environment.
Cosmic radiation
High energy rays from space which, in space, are primarily protons (92%)
and alpha particles (6%). In space, cosmic rays come from all directions.
DRAM
Dynamic RAM is typically the RAM used as the main
memory in a computer system. DRAM requires fewer transistors per bit than
SRAM, but each bit needs to be refreshed regularly
or it will lose information. DRAM is typically slower but much less expensive
than SRAM.
ECC
An error correcting code specifies how to add extra information to data
bits in a fashion that allows corrections to be made to the data if one
(or possibly more) of the resulting bits is changed. The most common error
correcting codes have the ability to correct a single bit error and detect
double bit errors. To add SEC-DED (single error correction - double error
detection) to a word of 64 bits, an 8 additional bits are usually used.
FIT
One Failure In Time corresponds to one fail per billion chip-hours.
Hamming code
Hamming codes are one of the most commonly used types of error correcting
codes.
Hard Error
A hard error is an error caused by a permanent physical defect in the
memory system.
HER
The hard error rate is the frequency of errors caused by permanent physical
defect in the memory system. The hard error rate is usually much lower than
the soft error rate.
MTBF
Mean time between failures...
Parity
Parity memory is used to detect memory errors. Each byte of data is
accompanied by a parity bit which is determined by the number of ones in
the eight data bits. Even (odd) parity ensures that the total number of
one bits in the data bits and parity bit is even (odd). Parity memory is
most commonly used on microcomputers with a small word size. A parity memory
system which uses a 64 bit word requires the same number of bits as
error correcting memory, which makes error correcting memory more appealing
for 64 bit and larger word sizes.
RAM
Random Access Memory (see also DRAM and
SRAM). Random access memory should allow
equally fast access to any memory location in the system.
Modern RAM systems are not quite random access, but compared to
disk drives, they provide a very good approximation to random access
memory. The term RAM, by itself, usually refers to the VLSI based
main memory of the computer system.
S-A-0
A memory cell exhibits a stuck-at-zero fault when its value cannot be
changed to the value 1.
S-A-1
A memory cell exhibits a stuck-at-one fault when its value cannot be
changed to the value 0.
Soft Error
A soft error is an error which is not due to any permanent physical
defect in the memory system. Soft errors can be fixed by either
writing new data to the invalid memory area or by restarting the computer.
SER
The soft error rate is the frequency of errors caused by
alpha particles, cosmic radiation)
and other factors which do not permanently damage the memory system.
SRAM
Static RAM is used for the cache memory and registers
in computer systems. SRAM typically requires four or six transistors per
bit, making it substantially more expensive than
DRAM, which usually requires one transistor per bit.
SRAM is able to operate at higher speeds than DRAM, and does not
require refreshing.
Transition fault
A transition fault is a fault in which a memory cell or line cannot
change from one particular state to a different state.
Yield
In integrated circuit manufacturing, the yield refers to the percentage
of usable dice obtained from a wafer. Lower yields result in higher cost
per chip, so increasing yield is an effective way of reducing the manufacturing
cost per chip.
