System Component Count is often one of the larger determining factors in system reliability. Let's compare two designs — two done largely with discrete and one done with ICs.
In the first version, the solution is a low integration modular, discrete-based design. This is for a transceiver with multiple bands handled via multiple LRUs (line replaceable units). In the second version, the integration is higher, but still largely discrete-based. Here, everything is in one LRU with much less wiring. In the last version, the system is largely a VLSI-based implementation that significantly reduces parts count from the first two approaches.
Version 1 has about 4,500 components, and very limited built-in test/built-in self-test (BIT/BIST) capability. Instead, it relies on pre/post dispatch checks. In this approach it is not unusual for wires to work loose and cause tuning errors or other faults in the system. With 4500 parts, even if each part were to only fail once per 100,000,000 hours, that works out to a failure about every two years.
Version 2 drastically reduces the wiring count, and adds BIT/BIST to the wire bundle. This is in addition to the pre/post dispatch manual checks. In this approach more emphasis can be given to the wiring due to the reduced number of interconnections and error checking added to the signaling over the wires. Undetected tuning or other errors are much less common.
With about 3,500 parts and the reduced wiring count for a basic system — again assuming each part were to fail only once per 100,000,000 hours — we could expect a fault approximately every 2.5 to 3.5 years.
In the VLSI version, the component count is reduced to about 450 parts. Error checking is prevalent in much of the system. There also error checking in the wiring, due to the reduced number of interconnections. Undetected errors are quite rare.
Using the same criteria and calculations as above in Versions 1 and 2, and considering the greatly reduced wiring + interconnections, there is a fault approximately every 10 to 20 years (some reduction is taken for the use environment and the SMD packaging). For the industrial environment, we use a value of 10 years = 100,000 hours of operation.
How does this compare with past generations of products you have worked with? Do VLSI/high integration solutions improve reliability in your situation? What are your experiences with reducing component count to improve reliability?