Translinear Circuits and Gilbert-Cell Multipliers
Introduction
Translinear circuits are a class of analog circuits that perform multiplication, division, and other algebraic operations on currents by exploiting the precise exponential relationship between current and voltage in a forward-biased junction. The name, coined by Barrie Gilbert in 1975, combines "transconductance" and "linear," reflecting the fact that the transconductance of a bipolar transistor is linearly proportional to its collector current. This proportionality is the hinge on which the entire technique turns: it converts the awkward exponential device characteristic into a clean, temperature-stable algebraic relationship among currents.
The approach is powerful because it computes with the physics of the device rather than against it. Most analog design treats the exponential current-voltage law of a junction as a nonlinearity to be confined to a small signal region or linearized away. Translinear design does the opposite, arranging junctions in closed loops so that the exponentials combine to yield an exact product or quotient of currents. The result is a family of compact, wideband circuits, the most celebrated of which is the Gilbert cell, the four-quadrant analog multiplier that became a standard component for mixing, modulation, gain control, and computation.
This article develops the translinear principle, shows how loops of junctions implement arithmetic on currents, and then examines the Gilbert cell in its principal roles as a multiplier, a mixer, and a variable-gain amplifier. It closes with the temperature behavior that makes the technique both robust and, in certain respects, demanding.
The Exponential Device Law
The foundation of translinear design is the relationship between the collector current of a bipolar transistor and its base-emitter voltage. To good approximation, the collector current grows exponentially with the base-emitter voltage, scaled by a saturation current that characterizes the device and divided in the exponent by the thermal voltage. The thermal voltage equals the product of Boltzmann's constant and absolute temperature divided by the electron charge, and at room temperature it is approximately 25.9 millivolts near 300 kelvin, often quoted as about 26 millivolts.
Two consequences of this law matter for what follows. First, because the relationship is exponential, the base-emitter voltage is the logarithm of the collector current, scaled by the thermal voltage. Logarithms turn products into sums, which is precisely why a loop of junctions can multiply. Second, the small-signal transconductance, the rate at which collector current changes with base-emitter voltage, is equal to the collector current divided by the thermal voltage. The transconductance is therefore directly proportional to the current, and this is the linear-in-current property from which the technique takes its name.
The same logarithmic behavior is available in a MOS transistor operated in weak inversion, also called the subthreshold region, where the drain current depends exponentially on the gate-source voltage. Translinear circuits can therefore be built in CMOS as well as in bipolar technology, although bipolar transistors generally provide closer adherence to the ideal exponential law over a wider range of currents.
The Translinear Principle
The translinear principle governs any closed loop of forward-biased junctions and states a remarkably simple result. Consider a loop traversed in one direction around the circuit, and divide the junctions in the loop into those whose polarity is encountered in the clockwise sense and those encountered in the counterclockwise sense. The principle asserts that, provided the loop contains an equal number of clockwise and counterclockwise junctions, the product of the current densities of the clockwise junctions equals the product of the current densities of the counterclockwise junctions.
The derivation is short. Around any closed loop the sum of the junction voltages must be zero by Kirchhoff's voltage law. Each junction voltage is the thermal voltage multiplied by the logarithm of its current density relative to its saturation current density. Setting the signed sum of these logarithmic voltages to zero, the thermal voltage cancels because it is common to every term, and exponentiating the resulting sum of logarithms converts it into an equality of products. The clockwise current densities, gathered on one side, equal the counterclockwise current densities on the other.
Two features of this statement deserve emphasis. The thermal voltage cancels out entirely, so the relationship among the currents is, to first order, independent of temperature, even though each individual junction voltage depends strongly on temperature. And if all junctions are made identical in area, the current densities are proportional to the currents themselves, so the principle reads directly as an equality of products of currents. This is the engine of translinear computation.
From Loops to Products and Quotients
A loop with two junctions on each side gives an equality of the form: one product of two currents equals another product of two currents. If three of those currents are inputs and the fourth is the output, solving the equality yields the output current as the product of two inputs divided by the third. A single translinear loop thus computes a product divided by a current, which is to say it multiplies and divides at once. By choosing which currents are forced and which are free, the same loop can be configured as a multiplier, a divider, a squarer, or a square-root extractor.
The squaring and square-rooting operations follow immediately. If two of the input currents are made equal, the output becomes proportional to the square of that current divided by the reference; conversely, arranging the loop so that the output appears on both sides forces the output to be the geometric mean, or square root, of the products of the other currents. Higher-order functions, including root-mean-square converters and vector-magnitude circuits, are built by combining these elementary loops.
Current-Mode Multipliers and Dividers
Translinear loops give rise to a broad family of current-input, current-output arithmetic blocks. Because the inputs and outputs are currents, these circuits compose naturally: the output current of one block becomes an input current of the next, summed at a node by Kirchhoff's current law without intervening voltage buffers. This composability, together with the inherent temperature cancellation, made translinear circuits attractive for analog computation long before digital signal processing became inexpensive.
The One-Quadrant Core
The simplest translinear multiplier handles unipolar currents only, operating in a single quadrant where all currents are positive. A four-junction loop relates an output current to two input currents and a reference current, producing the product of the two inputs divided by the reference. The reference current sets the scale and the units, and it is what makes the otherwise dimensionally awkward product of two currents come out as a current again.
This one-quadrant core is the seed from which the full four-quadrant multiplier grows. By itself it is useful for processing inherently positive quantities, such as the magnitudes used in automatic gain control or in power and root-mean-square measurement, where the sign of the signal is not required.
Dividers and Other Functions
Reconfiguring the same loop so that the quantity of interest appears in a denominator yields a divider. Analog dividers are valuable for ratiometric measurements, for normalization, and for implementing automatic level control, where a signal must be scaled by the reciprocal of a measured amplitude. Cascades and combinations of translinear loops also realize logarithmic and exponential transfer functions, trigonometric approximations, and the geometric mean, giving the designer a small algebra of currents from which to assemble more complex functions.
The Gilbert Cell
The Gilbert cell, introduced by Barrie Gilbert in 1968, is the most important translinear circuit and one of the most widely used analog building blocks of any kind. In its canonical form it is a four-quadrant analog multiplier, meaning that it correctly handles inputs of either sign and produces an output whose sign follows the product. Structurally it consists of cross-coupled differential pairs: an upper set of two emitter-coupled pairs whose outputs are interleaved, steered by the tail currents of a lower differential pair. One input controls the lower pair and thereby the bias currents of the upper pairs, while the other input controls the upper pairs directly.
The cell's operation can be understood through the translinear principle and through the transconductance relationship. The lower pair converts one input into a difference of bias currents. Those currents set the transconductance of the upper pairs, and because transconductance is proportional to current, the gain that the upper pairs apply to the second input is itself proportional to the first input. The cross-coupled connection of the upper outputs subtracts the two halves so that the differential output current is proportional to the product of the two inputs. The four-quadrant behavior arises naturally because either differential input can swing positive or negative about its common-mode point.
As an Analog Multiplier
Used as a precision multiplier, the Gilbert cell is augmented to make the product accurate and well defined. Predistortion circuits, typically logarithmic or diode-connected stages at the inputs, compensate for the nonlinearity of the differential pairs so that the transfer characteristic is a faithful product over a usefully large input range rather than only for small signals. A reference current or voltage sets the multiplier's scale factor. Integrated four-quadrant multipliers built on this principle have long been available as standard parts and are used for computation, modulation, phase detection, and the realization of squarers, frequency doublers, and root-mean-square converters.
As a Mixer
In radio-frequency systems the Gilbert cell is ubiquitous as an active mixer. Mixing is multiplication: applying a signal to one input and a local-oscillator waveform to the other produces an output containing sum and difference frequencies, which is exactly the frequency translation a mixer performs. When the local-oscillator drive is large enough to switch the upper pairs fully, the cell behaves as a doubly balanced mixer, in which both the signal and the local oscillator are balanced out of the output, leaving predominantly the wanted mixing products. This balance suppresses the local-oscillator feedthrough and even-order distortion that plague simpler mixers, which is why the Gilbert-cell mixer is a mainstay of integrated transceivers. Its conversion gain, rather than the loss of a passive mixer, and its modest local-oscillator power requirement are additional practical attractions.
As a Variable-Gain Amplifier
If one input of the Gilbert cell is a control signal and the other is the signal to be amplified, the cell becomes a variable-gain or voltage-controlled amplifier. The control input sets the bias currents and hence the transconductance applied to the signal, so the gain varies smoothly and continuously with the control. This makes the Gilbert cell the heart of many automatic-gain-control loops, of voltage-controlled amplifiers in instrumentation and audio, and of the linear-in-decibel gain stages used in receivers, where a control voltage that is linear in decibels is obtained by suitable shaping of the control path.
Temperature Behavior
Temperature occupies a curious dual role in translinear circuits, and understanding it is essential to using them well. The central virtue of the translinear principle is that the thermal voltage cancels in any balanced loop, so the algebraic relationship among the currents, the product, the quotient, the square root, is to first order independent of temperature. A translinear multiplier or divider therefore holds its defining current relationship remarkably steady as the chip warms or cools, a robustness that discrete analog computers built from individual components could never match.
This cancellation, however, is not unconditional, and several temperature effects remain. The most direct is the transconductance of a differential pair, which is inversely proportional to absolute temperature because the thermal voltage appears in the denominator. In a Gilbert cell whose gain is set by a fixed control current, the small-signal gain therefore drifts with temperature unless the bias is made proportional to absolute temperature. Designers commonly cancel this drift by deriving the bias current from a circuit whose output is proportional to absolute temperature, so that the temperature dependence of the bias offsets that of the transconductance and the overall gain becomes stable.
Two further effects limit precision. The saturation current of a junction is strongly temperature dependent, so any mismatch in saturation current between supposedly identical devices introduces an error that itself varies with temperature; tight device matching and good thermal coupling on the die are the remedies. And the input offsets and base currents of real transistors drift with temperature, degrading the accuracy of the product near zero. In careful designs these residual effects are kept small, so that the translinear circuit delivers an arithmetic operation on currents that is accurate and stable across a wide temperature range, while in less careful designs they set the practical limits on linearity and dc accuracy.
Practical Considerations and Limitations
Translinear circuits are not a universal solution, and their strengths come with constraints that shape where they are used. The exponential law that makes them work is an idealization. At high currents the series resistance of the emitter and the base spreads the device away from the ideal exponential, compressing the transfer characteristic, while at very low currents leakage and recombination corrupt the relationship. The usable range of currents over which the principle holds accurately is therefore bounded at both ends, and the designer must place the operating currents within that window.
Matching is the other dominant concern. Because the principle equates products of currents in nominally identical junctions, any mismatch in junction area or saturation current translates directly into a multiplicative error. Integrated implementations exploit the excellent matching of devices fabricated together on one die, with common-centroid layout and careful thermal symmetry, to keep these errors small. Bandwidth, by contrast, is a relative strength: the signal path is a low-impedance current path through forward-biased junctions, so translinear circuits and Gilbert cells are capable of operating to very high frequencies, which is one reason the Gilbert-cell mixer is so prevalent in radio-frequency integrated circuits.
Summary
Translinear circuits perform exact algebraic operations on currents by arranging forward-biased junctions in closed loops and exploiting the exponential current-voltage law of those junctions. The translinear principle states that, in a balanced loop, the product of the current densities of the junctions oriented one way equals the product of those oriented the other way, with the thermal voltage canceling so that the result is to first order independent of temperature. From this single principle follow current-mode multipliers, dividers, squarers, and square-root and root-mean-square circuits, all of which compose naturally because currents sum at a node.
The Gilbert cell is the preeminent translinear circuit, a four-quadrant analog multiplier built from cross-coupled differential pairs whose transconductance is proportional to a controlling current. The same cell serves as a precision multiplier when linearized by input predistortion, as a doubly balanced active mixer that translates frequency while suppressing local-oscillator feedthrough, and as a variable-gain amplifier whose gain tracks a control signal. Temperature both helps and challenges the technique: the thermal voltage cancels in the defining current relationship, but the gain of a differential pair varies inversely with absolute temperature and is commonly stabilized by a proportional-to-absolute-temperature bias. Bounded current range and device matching set the practical limits, but within them translinear design yields compact, wideband, and temperature-robust analog computation and signal processing.
Related Topics
- Current Conveyors - Current-mode building blocks whose voltage-follower and current-mirror cores rest on the same translinear loops.
- Current-Feedback Amplifiers - Another current-mode topology that, like translinear circuits, exploits low-impedance current paths for wide bandwidth.
- Operational Amplifiers and Linear Circuits - The linear-amplifier counterpart against which nonlinear translinear computation is contrasted.
- Modulation and Demodulation - Mixing and modulation, a principal application of Gilbert-cell multipliers.
- RF and High-Frequency Analog - The radio-frequency domain in which the doubly balanced Gilbert-cell mixer is a mainstay.
- Advanced Analog Concepts - The parent category collecting current-mode and physics-as-computation analog techniques.