Low-Power Regenerative Amplifier

• Low-Power Regenerative Amplifier is a system that exploits recirculating feedback and nonlinear effects to achieve substantial gain at minimal power consumption.
• It employs high-Q microresonators in optical systems and regenerative current mirrors in CMOS technology to ensure stable, low-noise amplification.
• These devices are critical for applications in on-chip photonics, LIDAR, advanced imaging, and signal processing where compactness and efficiency are paramount.

A low-power regenerative amplifier is a class of amplification system in which positive feedback and nonlinear interaction mechanisms are exploited to yield substantial gain at minimal power consumption. In both photonic and electronic domains, regenerative amplifiers operate by recirculating signal energy or current in feedback-enabled structures, where carefully tuned loop dynamics and element nonlinearities ensure high gain, low noise, and small physical and electrical footprints. Two major exemplars are the microresonator-assisted regenerative optical-parametric amplifier (OPA) for on-chip photonics (Zhao et al., 2022) and the regenerative current mirror for CMOS front-end signal processing (Wu, 2023). These distinct platforms implement regenerative principles to suit their respective operational regimes, but both share core architectural advantages relevant for ultralow-power, high-density, and low-noise applications.

1. Regenerative Amplification Principles

In a microresonator-assisted OPA, regenerative gain arises from the phase-matched four-wave mixing (FWM) process within a high-Q cavity. The device is pumped by a continuous-wave laser at angular frequency $\omega_p$, inducing the simultaneous generation of signal ($\omega_s$) and idler ($\omega_i$) sidebands such that $2\omega_p = \omega_s + \omega_i$. The microresonator's high field buildup enhances the FWM process and provides recirculating feedback: each round-trip, a weak input signal is further amplified provided that the small-signal parametric gain surpasses total cavity losses. This closes a "regenerative loop," enabling net gain directly on-chip without auxiliary amplification stages (Zhao et al., 2022).

In the electronic domain, the regenerative current mirror topology exploits positive feedback between dual current-mirror blocks (NMOS and PMOS). Both mirrors are configured with open-loop gain nominally less than unity. When cross-coupled, the loop gain can approach (but not reach) unity, producing large closed-loop gain even with extremely low bias currents. The positive feedback effectively boosts gain while avoiding instability, thereby enabling ultralow power operation in front-end amplifier stages (Wu, 2023).

2. Mathematical Frameworks and Gain Formulation

The microresonator OPA model is rooted in mean-field coupled-mode equations for the cavity amplitudes ($a_p$, $a_s$, $a_i$). The Hamiltonian-form dynamics, with Kerr nonlinearity $\gamma = (n_2 \omega_p)/(c A_\text{eff})$ and total loss rates $\kappa_j$, are

$$\frac{d a_p}{d t} = -\Bigl(\tfrac{\kappa_p}{2}+i\Delta_p\Bigr)a_p - i\gamma\Bigl( |a_p|^2 + 2|a_s|^2 + 2|a_i|^2\Bigr)a_p - i\gamma a_s a_i a_p^* + \sqrt{\kappa_{c,p}} s_{p,\rm in}$$

$$\frac{d a_s}{d t} = -\Bigl(\tfrac{\kappa_s}{2}+i\Delta_s\Bigr)a_s - i2\gamma( |a_p|^2 + |a_s|^2 + |a_i|^2 )a_s - i\gamma a_p^2 a_i^* + \sqrt{\kappa_{c,s}} s_{s,\rm in}$$

$$\frac{d a_i}{d t} = -\Bigl(\tfrac{\kappa_i}{2}+i\Delta_i\Bigr)a_i - i2\gamma( |a_p|^2 + |a_s|^2 + |a_i|^2 )a_i - i\gamma a_p^2 a_s^*$$

Linearization for small signal yields gain per round trip:

$$G(\omega) = 1 + \frac{4\gamma^2 P_p^2 Q_p^2 Q_s Q_i}{\left[ (\omega-\omega_s+\delta_s)^2 + (\kappa_s/2)^2 \right] \left[ (\omega-\omega_i+\delta_i)^2 + (\kappa_i/2)^2 \right] }$$

The oscillation threshold (parametric oscillation) occurs at pump power

$$P_\text{th} = \frac{\hbar\omega_p \kappa_p^2}{16 \gamma Q_s Q_i \eta_c}\!, \quad \eta_c = \frac{\kappa_{c,p}}{\kappa_p}$$

For the regenerative current mirror (RCM), the translinear gain structure is:

For each unidirectional current mirror (NMOS/PMOS) with matched devices $(g_m, r_o)$,

$A_\text{NMOS} = -\frac{g_m r_o}{1 + g_m r_o}$

$$A_\text{PMOS} = -\frac{g_{mp} r_{op}}{1 + g_{mp} r_{op}}$$

The full loop gain is $A_\text{loop} = A_\text{NMOS} A_\text{PMOS}$ (positive sign). The closed-loop gain is

$$G_\text{CL} = \frac{A_\text{loop}}{1 - A_\text{loop}}$$

By setting device overdrive and degeneration, $A_\text{loop}$ is maintained below but near unity for high gain and stability.

3. Physical Implementation and Typical Performance

Table 1: Comparative Physical and Performance Parameters

System	Material/Tech	Power/Channel	Gain	Bandwidth	Footprint
Microres. OPA	Si₃N₄ microring, SiO₂	$< 20$ mW (total)	up to 30 dB	1–2 THz	$\sim$0.002 mm²
RCM (65 nm)	CMOS (NMOS/PMOS)	$< 10\ \mu$W	$5{-}10\times$	Tens of MHz	Standard CMOS cell

Microresonator OPAs employ radii of 20–30 μm integrated on Si₃N₄ waveguides (thickness ~600 nm, width ~1.2 μm), with SiO₂ cladding. Intrinsic Q-factors are $(1{-}3)\times10^6$; coupling Q is $(0.5{-}2)\times10^6$. Nonlinear index $$n_2(\text{Si}_3\text{N}_4) \simeq 2.4\times10^{-19}$$ m²/W, $A_\text{eff}\sim 1\ \mu\text{m}^2$, yields $\gamma\sim 1–2$ W⁻¹ m⁻¹. Demonstrated small-signal gain reaches 30 dB at only $\sim$9 mW on-chip pump power, with 3 dB gain bandwidth of 10–20 nm. Power budget for all amplifier operation, including auxiliary heaters, remains below 20 mW (Zhao et al., 2022).

The regenerative current mirror, implemented in 65 nm CMOS, uses bias currents as low as 1 μA per mirror column, with total power consumption per channel below 10 μW. Transient current gain up to 5× was realized, and two-stage amplifier chains yielded CMOS logic level outputs ($\sim$1.2 V) from sub-fC input charges in less than 300 ns (Wu, 2023).

4. Dispersion Control, Phase Matching, and Loop Gain Optimization

Phase-matching in the OPA is governed by both geometric and nonlinear contributions. The generalized phase-mismatch is

$$\Delta k(\Omega) = 2k_p - k_s - k_i + 2\gamma P_p \approx \beta_2 \Omega^2 + \frac{1}{6}\beta_3 \Omega^3 + 2\gamma P_p = 0$$

Here, group-velocity dispersion ($\beta_2$) and third-order dispersion ($\beta_3$), as well as pump-induced nonlinear shifts, must be engineered for broadband gain. Coupled cavity structures (e.g., dual rings or pulley-sectioned waveguides) enable flattening of $\beta_2$ over broader spans, increasing gain bandwidth by factors of 2–3 (Zhao et al., 2022).

In electronic regenerative current mirrors, loop gain must be optimized to maximize amplification without risking oscillatory or latching instability. This is achieved by biasing NMOS/PMOS devices such that $A_\text{NMOS}<1$, $A_\text{PMOS}<1$, and $A_\text{loop}=A_\text{NMOS}\cdot A_\text{PMOS}<1$ in all corners. An additional series degeneration device (e.g., M15, gate controlled by $V_\text{B1}$) allows precise adjustment of open-loop gain to stay below criticality. Monte Carlo verification across process variations ensures safety margin (recommended 5–20%) (Wu, 2023).

5. Regimes of Operation: Thresholds, Injection Locking, and Noise

OPA-based regenerative amplifiers exhibit bifurcated operation below and above the oscillation threshold $P_\text{th}$. Below threshold, externally injected signals are amplified; above threshold, spontaneous parametric oscillation (OPO) occurs—spontaneously growing signal/idler pairs seeded by vacuum noise, saturating when parametric gain equals cavity loss:

$4\gamma^2 P_p^2 Q_s Q_i = \kappa_s \kappa_i$

With weak external seeding, injection locking stabilizes the OPO to the input frequency, suppressing additional sidebands and effectively eliminating excess noise. The amplifier attains maximum gain, e.g. 30 dB, without further increase in $P_p$ once the seed power exceeds $\sim$1 μW, offering a regime of near-quantum-limited noise figure (NF $\sim$3–4 dB) and dynamic range spanning 1 nW–1 μW (Zhao et al., 2022).

For regenerative current mirrors, the closed-loop gain is maximized with loop gain near but below unity. Output linearity is ultimately limited by the maximum gain sustainable without crossing into oscillation. Simulations indicate a small-signal bandwidth of several tens of MHz, with full-scale rail-to-rail logic swing in cascaded amplifier-buffer stages. Input-referred noise performance is coupled to device sizing and bias; explicit noise spectra were not provided (Wu, 2023).

6. Implementation Considerations and Application Domains

Optimization of regenerative amplifiers in either platform involves trade-offs among gain, stability, noise, and power. In OPA designs, increasing $Q_i$ (e.g., above $10^7$) and reducing $A_\text{eff}$ (slot-waveguide approaches) feasibly lowers threshold $P_\text{th}$ below 1 mW and extends operation to nanowatt signal levels. Dispersion engineering is critical for application-specific spectral tailoring (Zhao et al., 2022).

In CMOS regenerative current mirrors, loop gain must remain below unity under all PVT (process, voltage, temperature) conditions; the degeneration control (e.g., $V_\text{B1}$ for M15) must be dimensioned to guarantee this. Device sizing navigates between low-noise and fast-settling response—wider devices for higher $g_m$, balanced against parasitic capacitance. Larger arrays (e.g., high-energy physics trackers) benefit from per-column or per-bank digital trim of degeneration for post-fabrication compensation.

Both architectures excel in environments where ultralow electrical power, high density, and compactness are critical. Recognized application domains include portable and space-constrained coherent LIDAR, on-chip optical and microwave-frequency synthesis, advanced timekeeping systems (Zhao et al., 2022), as well as very-large-format silicon pixel detectors deployed in high-energy physics tracking and advanced medical imaging (Wu, 2023). The use of regenerative architectures is a principal enabler for extending signal amplification into previously inaccessible ultralow-power regimes.