Deep Submicron CMOS Scaling 1990s Weigh Oxide Thinning Against Reliability and Leakage
# Deep Submicron CMOS Scaling (1990s): Step 6 — Weigh Oxide Thinning Against Reliability and Leakage Tradeoffs
## 1. Combining Two Independent Field-Driven Risks
Steps 2 through 4 addressed hot-carrier injection by grading the drain field; Step 5 addressed gate control by thinning the oxide. These two levers interact: a thinner gate oxide raises the oxide's own electric field for a given gate voltage, and a higher oxide field independently raises the probability that any carrier that does make it into the oxide — whether via tunneling or hot-carrier injection — creates lasting damage. A 1990s process could not optimize channel doping and oxide thickness in isolation from each other; both had to be set together against a shared reliability budget.
## 2. Time-Dependent Dielectric Breakdown as the Combined Risk
Beyond the two mechanisms already covered, a thinner oxide under sustained voltage stress is also more susceptible to time-dependent dielectric breakdown — a slow accumulation of defects through the oxide bulk under normal operating field, distinct from both hot-carrier injection (localized near the drain) and direct tunneling (present everywhere, not cumulative). Qualifying a 1990s process generation required separately characterizing all three mechanisms — hot-carrier lifetime, static tunneling leakage, and time-dependent breakdown lifetime — against the chosen combination of oxide thickness, supply voltage, and LDD grading, because none of the three could be inferred safely from the others.
## 3. Why This Made Process Qualification Slower, Not Faster
Each new process generation's qualification suite grew to include accelerated stress tests for all three mechanisms at multiple voltage and temperature points, because the field-vs-voltage lag from Step 1 meant that a thickness/voltage/doping combination that passed one mechanism's reliability target could still fail another's. This is part of why deep submicron process generations in the 1990s took longer and required more characterization effort per node than the simpler geometric shrinks of the 1980s, even though the device topology itself (planar CMOS, polysilicon gate) had not fundamentally changed.
## 4. The Resulting Design Rule
The practical outcome was a set of coupled design rules — maximum allowed oxide field as a function of required hot-carrier lifetime and required breakdown lifetime, jointly constraining the allowable combinations of oxide thickness and supply voltage for a given channel length. These coupled rules, rather than any single physical limit, are what ultimately bounded how aggressively a 1990s process could scale oxide thickness down at a given voltage — directly motivating the voltage-scaling steps later in this series (17 and 18), where lowering supply voltage buys back margin on all three reliability mechanisms simultaneously.