Threshold Voltage
Threshold voltage (Vth) is the minimum gate-to-source voltage required to create a conducting channel between the source and drain terminals of a MOSFET transistor. Below the threshold voltage, the transistor is off — the channel is depleted of charge carriers, and current flow is negligible. Above the threshold, the channel inverts, and the transistor can conduct. This simple binary — on or off, conducting or insulating — is the foundation of digital logic. Every bit stored, every gate computed, every instruction executed depends on transistors switching reliably across the threshold.
The threshold voltage is not an arbitrary design parameter. It is determined by the physical properties of the transistor: the doping concentration of the channel, the thickness of the gate oxide (or the equivalent oxide thickness in high-κ dielectrics), the work function of the gate material, and the geometry of the device. To make transistors switch faster — to build the smaller, faster chips that Moore's law promised — engineers reduced the gate oxide thickness and the channel length, which in turn lowered the threshold voltage. Lower Vth means faster switching, because the gate needs less voltage to turn the channel on. But lower Vth also means higher leakage current when the transistor is off.
The Subthreshold Leakage Crisis
When a transistor is "off" — gate voltage below threshold — it is not perfectly off. A small leakage current still flows between source and drain due to thermionic emission and quantum tunneling. This subthreshold leakage current is exponentially sensitive to threshold voltage: Ileak ∝ exp(−Vth/nVT), where VT is the thermal voltage (about 26 mV at room temperature) and n is a device parameter. A reduction of Vth by 60 mV increases leakage current by a factor of ten.
For decades, this exponential relationship was manageable because leakage was a small fraction of total chip power. Dynamic power — the power consumed when transistors switch — dominated. But as Vth was scaled down to maintain performance while supply voltage dropped, leakage power grew exponentially. Around the 65 nm technology node (circa 2005), leakage power became comparable to dynamic power in many designs. By 22 nm, leakage was a first-class design constraint.
The industry responded with a series of techniques: high-κ dielectrics to reduce gate leakage, multi-threshold designs (mixing high-Vth and low-Vth transistors on the same die), power gating (physically cutting power to idle blocks), and body biasing (dynamically adjusting Vth by manipulating the substrate voltage). These techniques mitigated the problem but did not solve it. The fundamental trade-off remains: lower Vth for speed, higher Vth for leakage control.
The End of Dennard Scaling
Dennard scaling — the observation that transistor dimensions, voltage, and current could all be scaled proportionally, keeping power density constant — assumed that Vth could be scaled down along with supply voltage. This assumption held for roughly thirty years. But Vth cannot be scaled arbitrarily low. At room temperature, Vth cannot practically go below 200-300 mV without catastrophic leakage. Below this, the transistor cannot distinguish between "on" and "off" with sufficient noise margin.
When Dennard scaling broke down, the industry was forced to continue reducing transistor dimensions (to increase density) without the corresponding reduction in voltage (to maintain performance). The result was the power wall: power density increased with each generation, and chips could no longer be fully powered without exceeding thermal limits. The portion of the chip that must remain dark — dark silicon — is a direct consequence of the threshold voltage floor. We cannot scale Vth further, so we cannot scale voltage further, so we cannot power all transistors simultaneously.
Threshold Voltage and the Limits of Computing
The threshold voltage is a physical constant, not an engineering variable. It is bounded below by thermal noise (kT/q ≈ 26 mV at room temperature) and by the fundamental requirement that a digital device must have a noise margin — a gap between the "on" and "off" states that is large enough to be immune to noise, process variation, and aging. At the limit, a single-electron transistor or a quantum-dot device could in principle operate with sub-thermal thresholds, but these devices face their own scaling challenges and are not yet practical for general-purpose logic.
Some researchers propose operating circuits at cryogenic temperatures to reduce thermal voltage and allow lower Vth. This is viable for specialized high-performance computing (some supercomputers already cool processors to liquid nitrogen temperatures) but is not a general solution for consumer or mobile devices. Others propose moving beyond the MOSFET entirely — to tunnel FETs, negative capacitance FETs, or spintronic devices — that exploit quantum effects to achieve steeper subthreshold swings. These are active research directions but have not yet displaced silicon CMOS.
The threshold voltage, in its stubborn refusal to scale, reveals that the end of Moore's law is not a marketing problem or a temporary engineering challenge. It is a physical boundary. The transistors we build are made of atoms, and atoms have a size. The voltages we apply are made of electrons, and electrons have a charge and a temperature. Threshold voltage is where these fundamental constants meet the engineering of computation. The wall is real.