Subthreshold computing (also called sub-Vth or near-threshold computing) operates transistors at supply voltages below the threshold voltage ($V_{th}$) — exploiting the weak inversion (subthreshold) current regime to achieve ultra-low power consumption at the cost of dramatically reduced speed.

How Subthreshold Operation Works

• Normally, transistors switch between "off" (below $V_{th}$) and "on" (above $V_{th}$) — logic operates in strong inversion.
• In subthreshold computing, $V_{DD} < V_{th}$ — transistors never fully "turn on." They operate in weak inversion where current is exponentially dependent on gate voltage:

Where $V_T = kT/q \approx 26$ mV at room temperature and $n$ is the subthreshold swing factor.

Power Savings

• Dynamic Power: $P_{dyn} \propto V_{DD}^2$. At $V_{DD} = 0.3V$ vs. 1.0V: power is reduced to ~9% — over 10× savings.
• Total Energy per Operation: Energy = Power × Delay. Even though delay increases dramatically, the energy per operation still decreases significantly in subthreshold — there is an energy-optimal voltage (typically 0.3–0.4V for modern processes).
• Leakage: At subthreshold voltages, leakage power becomes comparable to or even dominates dynamic power — the crossover point.

Performance Impact

• Speed: Subthreshold circuits are 100–1000× slower than nominal-voltage operation. Clock frequencies drop from GHz to MHz or even kHz.
• Delay Variability: In subthreshold, delay is exponentially sensitive to $V_{th}$ variation — process variation causes huge delay spread (10× or more between fast and slow devices).
• This means subthreshold computing is only viable for applications where speed is not critical but power is paramount.

Applications

• IoT Sensors: Wireless sensor nodes that wake periodically, sample data, and transmit — compute at kHz–MHz rates, must last years on a coin cell battery.
• Biomedical Implants: Pacemakers, neural interfaces, hearing aids — ultra-low power, very low data rates.
• Wearables: Activity trackers, environmental monitors — low compute needs, small batteries.
• Energy Harvesting: Devices powered by solar, thermal, or RF energy harvesting — available power is microwatts to milliwatts.

Design Challenges

• Variation Sensitivity: Exponential dependence on $V_{th}$ makes circuits extremely sensitive to process variation — requires robust design techniques (upsized transistors, body biasing, variation-tolerant architectures).
• Reduced Noise Margins: $V_{DD}$ is small, and the voltage swing between logic 0 and 1 is tiny — susceptibility to noise increases dramatically.
• Standard Cells: Conventional cell libraries are not optimized for subthreshold — dedicated subthreshold standard cell libraries with larger transistors and different topologies are needed.
• SRAM Stability: SRAM is particularly challenging at subthreshold — read and write stability degrade significantly. Specialized bit cell designs (8T, 10T) are required.
• Minimum Energy Point (MEP): The optimal $V_{DD}$ where total energy (dynamic + leakage) is minimized — depends on the specific technology and workload.

Subthreshold computing represents the extreme end of low-power design — it trades speed for extraordinary energy efficiency, enabling applications that would be impossible with conventional voltage operation.