DC coupled driver with active termination

[0001] The present invention claims priority from U.S. Patent Application No. 61/047,981 filed Apr. 25, 2008, entitled “DC Coupled Driver With Active Termination”, by Nelson, and claims priority from U.S. Patent Application No. 61/104,480 filed Oct. 10, 2008, entitled “DC Coupled Driver With Active Termination”, by Nelson, which are incorporated herein by reference for all purposes.

[0002] The present invention is directed to driver electronics, and more particularly, to integrated circuit drivers for opto-electronic emitters such as vertical cavity surface emitting lasers.

[0003] In applications where optical signals are used to transmit data, such as fiber-optic communications systems, fast and efficient driver circuits are required for modulating the optical emission of optical transmitters such as laser diodes, light emitting diodes, vertical cavity surface emitting lasers, and similar opto-electronic devices. Where the modulation rates extend into regions where electrical reflections within interconnections between the driver circuitry and the driven opto-electronic device cannot be neglected, electrical terminations are often required.

[0004] A commonly used driver arrangement in prior art is a DC coupled driver 100 for driving a VCSEL D as shown in FIG. 1. It comprises a differential input npn bipolar transistor pair q₃ and q₄. Differential voltage inputs v_in and v_ip are connected to a base on the transistors q₃ and q₄ to excite collector currents i₃ and i₄, respectively. A supply voltage V_dd is connected to a collector of the transistor q₄ and through a current source I to a collector of the transistor q₃. An anode of the VCSEL D receives a voltage V_d at the collector of the transistor q₃ to produce a driving current i_D. A cathode of the VCSEL D is connected to a circuit ground. Emitters of the transistors q₃ and q₄ are also connected to the circuit ground, but through a current source 2M. There is no back-termination.

[0005] The output current range is I for a ‘one’ level and I−2M for a ‘zero’ level. This prior art driver has a power dissipation
P_diss=V_dd*(I+M).  Equation [1a]

[0006] This driver has a problem in that without the back termination the circuit is susceptible to signal degradation due to reflections resulting from wire-bonds which are typically used to provide a connection between the driver output V_d and the load, VCSEL D. Even when the wire-bond length is held to small values like 1 mm or less, the reflection amplitude and delay can cause significant degradation in an eye diagram on the driver output V_d. To prevent this reflection from causing severe problems, the rise and fall time of the transistors q₃ and q₄ must be increased, thus precluding operation at very high modulation rates.

[0007] Another prior art driver 200 shown in FIG. 2 results when a passive termination resistor P is connected between the collectors of the transistors q₃ and q₄. For greatest effect, the value of the passive termination resistor P is adjusted to equal a VCSEL impedance, z_D. In addition, an upper current source 2M′ is inserted between the collector of the transistor q₄ and the voltage supply V_dd. A by-pass capacitor C_b is provided between the circuit ground and the collector of the transistor q₄.

[0008] In this circuit, the average bias in VCSEL D is set by the current source I, rather than the ‘one’ level. The ‘one’ level consists of a current (I+M) and ‘zero’ level consists of a current (I−M). While this circuit is fast and provides good eye quality, its power dissipation is higher than for the driver 100 in FIG. 1:
P_diss=V_dd*(I+2M).  Equation [1b]

[0009] This driver circuit can maintain a nearly constant power supply current, with small transient spikes occurring at the data transitions.

[0010] However, it has a problem in that for a large current M and high VCSEL voltage or impedance, a typical 3.3V power supply voltage is insufficient, as too much voltage is dropped across the passive termination resistor P. That leaves too little voltage for the upper current source 2M′ to operate properly when the VCSEL voltage is high.

[0011] Another drawback stems from difficulties in handling long consecutive identical digit (CID) data streams. Since node V_r is supplied by a current source rather than a voltage source, its DC value can drift in the presence of CID. The most practical way to prevent this problem from being severe is to use large values of the by-pass capacitor C_b, but this is costly in terms of integrated circuit die area. Still, well designed circuits of this type may see ‘one’ or ‘zero’ levels shift by up to 10% in the presence of 50 CIDs.

[0012] To reduce the severity of this problem, the current source 2M can be augmented with a shunt amplifier to the circuit ground. However the DC value drift is reduced at the expense of higher power dissipation. A prior art driver 300 shown in FIG. 3 was designed to overcome the shortcomings of the driver 100 in FIG. 1 by removing constraints on the V_dd, as well as including an active back termination but with reduced power dissipation.

[0013] An additional differential bipolar transistor pair q₅, q₆ is driven by the same differential voltage inputs v_in and v_ip as the differential pair q₃, q₄, respectively, to create a voltage at node V₃ which tracks the voltage at V_d. The emitters of the transistors q₅, q₆ are connected to the circuit ground through an augmenting current source 2A. A unity gain follower x₁ buffers voltage V₃ and drives a resistor S, which provides the termination impedance. The value of the resistor S can be partially or fully provided by the inherent output impedance of unity gain follower x₁. The voltage generated across resistor T by current i₅ is chosen to track the voltage V_d such that V₃ and V_d track each other. Since V₃ and V_d are ideally always equal, no current flows through the resistor S. A bias loop including a low speed operational amplifier U4 drives a gate of a PMOS FET m_r to produce a voltage V_r at a drain of the PMOS FET m_r to ensure that the average current in S is near zero. The power dissipation of this circuit is:
P_diss=V_dd*(I+M+2A+i(x₁)),  Equation [1c]

[0014] where i(x₁) is the current drawn by unity gain follower x₁, I, M, 2A are currents in the current sources I, 2M, 2A and V_dd is a power supply voltage.

[0015] This is typically higher than for driver 100 of FIG. 1 but less than for driver 200 of FIG. 2.

[0016] As in the other prior art, the supply current drawn by the driver 300 is not constant, so possibilities of cross-talk and electromagnetic interference exist. The headroom to supply voltage V_dd is greater in this prior art driver 300 than that of the driver 200 in FIG. 2. Typically the voltage V_r is higher than the voltage V_d by a factor of T*A, which may be several hundred millivolts. Also, if the voltage difference V_dd−V_r drops below several hundred millivolts, the PMOS FET m_r will cease to function as a current source. Typically, these factors confine the supply voltage to V_dd>3V and the diode voltage to V_d<2.2V in the bipolar case, thereby limiting the operating range and selection of the VCSEL D, leading to a great loss in accuracy as these limits are neared and current sources start to saturate.

[0017] Another problem present in prior art driver 300 is that due to a group delay in the unity gain follower x1 and mismatched time constants at circuit nodes V₃ and V_d, the dynamic voltage at node V₄ always lags the voltage at node V_d. At 10 Gbps data rates with a 100 pS bit period, this lag, typically 10 ps, can be severe. It slows the rise and fall time at node V_d, since during slewing, the output drive current i_D is stolen because the current in resistor S does not equal zero.

[0018] Another problem with the prior art driver 300 stems from the voltage V_r being supplied from a current source comprising the PMOS FET m_r. Thus, if long CIDs are present in the input signal stream, the value of voltage V_r will drift with a bias loop time constant. The only way to prevent this problem from being severe is to use large values of by-pass capacitor C_b, but this is costly in terms of die area. Nonetheless, well designed circuits of this type may see ‘one’ or ‘zero’ levels shift by up to 5% in the presence of 50 CIDs. A shunt voltage regulator can also be added in parallel with the current source to overcome this, but at the expense of increased power dissipation.

[0019] In FIG. 4 a further example of a prior art VCSEL driver 400 is shown, which possesses a very low power dissipation given by:
P_diss=V_cc*I_m+V_dd*I_b.  Equation [1d]

[0020] Differential voltage inputs v_ip and v_in are each connected to a gate on a differential input NMOS field-effect transistor (FET) pair m_p and m_n, respectively, each of whose drain is connected to a first supply voltage V_cc through a load resistor R_c. Sources of the transistors m_p and m_n are connected together to a modulation current source I_m working into the ground connection. The cathode of a diode D (such as a VCSEL) is connected to the drain of transistor m_n, while its anode is driven by a bias current source I_b from a second supply voltage V_dd. The anode of the VCSEL D is AC grounded through a bypass capacitor C_g.

[0021] The differential voltage inputs v_ip and v_in are driven so as to switch all of current I_m into the load resistor R_c on the left or into the parallel combination of load resistor R_c on the right and the VCSEL diode D. The average or bias current in diode D is just the upper current source current I_b. At high modulation rates, where the bypass capacitor C_g acts like an AC short circuit, the current in diode D will be:
I_d=I_b±I_m*R_c/(z_D+R_c),  Equation [1e]

[0022] where z_D is the impedance of the diode D.

[0023] Baseline wander is a bias shift away from a statistical 50/50 ‘one’ level to ‘zero’ level balance. Problems occur with the VCSEL driver 400 when data fed to inputs v_ip, v_in does not maintain the 50/50 balance of ‘one’ level/‘zero’ in the long term, or where there are long sequences of CIDs that shift a DC balance over a relatively long period of time.

[0024] The time constant of this circuit is:
t_c=C_g*R_c∥z_D,  Equation [1f]

[0025] where ∥ denotes the parallel value of impedance formed by resistor R_c and diode impedance z_D. Since this forms a single pole response, then for a constant v_ip−v_in input voltage, the current through diode D will settle to the current I_b from the starting point of I_b±I_m*R_c/(z_D+R_c) after approximately 2.2 time constants. Consider an example where this VCSEL driver 400 is designed to operate at 10 Gbps.

[0026] Typically, to keep the baseline wander small, a design may specify that a low frequency −3 dB corner of the driver circuit 400 be less than 300 kHz. For the parallel value of impedance R_c∥z_D equal to 70 ohm, the value of C_g required to achieve this is:
C_g=1/((R_c∥z_D)*2π*300 kHz)=7.6 nF.  Equation [1g]

[0027] This amount of capacitance is much larger than a reasonable amount of capacitance available in a typical integrated circuit process. In many applications it is not feasible to use an off-chip capacitor because of the parasitics encountered in going off chip, or because there is no room for such a capacitor. Hence the 300 kHz low frequency corner cannot be achieved, and this circuit would not be usable for 10 Gbps or lower data rates where 2³¹−1 pseudo-random bit sequence (PRBS) data is present.

[0028] If, for example, even the large value of 0.1 nF of on-chip capacitance could be provided for C_g, then the time constant of this circuit would be 0.1 nF*70 ohms=7 ns. If the input data were a 2³¹−1 PRBS, where 31 CIDs are occasionally encountered, then such a CID event would last about 3.1 ns. In that amount of time the DC value of the current in diode D can shift by nearly 10%.

[0029] Additionally, the worst case imbalance occurs when the running digital sum reaches its maximum, in which case the DC shift is even greater. This amount of DC shift, or baseline wander could not be tolerated in a VCSEL driver application.

[0030] A final prior art common anode VCSEL driver 500 is shown in FIG. 5. The magnitude of the differential driving voltage |v_ip−v_in| should be large enough to allow complete switching of the modulation current i_m between transistors m_p and m_n. The anode of the VCSEL diode D is connected to a positive supply voltage V_dd, while its cathode is biased by a bias current source i_z from the circuit ground so that a minimum of the diode current i_D, or ‘zero’ level current is provided just above a threshold of the VCSEL diode D. When the input v_ip sufficiently exceeds the voltage at input v_in, all of the modulation current i_m is switched through transistor m_p, increasing the diode current i_D to i_m+i_z=i_o, or the ‘one’ level current.

[0031] When v_in exceeds v_ip and current i_m is switched through transistor mn, then the VCSEL current i_D=i_z. The VCSEL forward voltage drop typically ranges from 1.6 V at currents just above threshold to 2.2 V at higher currents typical of the one level current. Therefore, to allow approximately 1 V at the drain of transistor m_p, V_dd will typically be 3V or greater. V_cc may be equal to V_dd, or it could be lower, around 1 V, to reduce power dissipation and prevent large V_ds at transistor m_n. Under typical conditions where the duty cycle of v_ip and v_in is 50%, the power dissipation of the driver 500 is given by Equation 1h, which is typically the lowest power dissipation of any type of VCSEL driver.
P_diss=0.5*i_m*(V_cc+V_dd)+i_z*V_dd  Equation [1h]

[0032] This type of circuit 500 is also capable of operation at very high data rates. At data rates greater than 10 Gbps it is very important that the distance between the driver circuitry, typically located on an integrated circuit, be very close to the VCSEL diode which is separate from the driver integrated circuit and connected to it by means of a wire-bond or printed circuit trace on a printed circuit board. Keeping this distance short is important to ensure that reflections caused by impedance mismatches between the high output impedance of the VCSEL driver and the much lower impedance of the VCSEL can be attenuated quickly. If the distance between the two is large, then the reflections from the impedance mismatches will reverberate back and forth between the two ends of the wire-bond or printed circuit trace for long periods of time, potentially interfering with subsequent data values.

[0033] A DC coupled driver for driving an optoelectronic emitter such as a vertical cavity surface emitting laser (VCSEL) is presented, which overcomes the shortcomings of the described prior art.

[0034] It is an object of the invention to provide a DC-coupled driver for high speed modulation of a VCSEL with improved immunity to pulse reflections in the interconnect between the driver and the VCSEL by incorporating an active termination.

[0035] A further object of the invention is to define different integrated circuit technologies for the DC-coupled driver, which are suitable for large scale manufacturing, including npn bipolar transistor and CMOS (NMOS and PMOS) technologies.

[0036] Another object of the invention is to achieve a low total power dissipation for the driver.

[0037] Provision is also made for using the active termination to control rise and fall times of the driver by adding a time delayed component of the input signal.

[0038] A DC-coupled driver for a vertical cavity surface emitting laser (VCSEL) diode incorporating a differential input stage for receiving data and complementary data signals from a signal source. A termination resistor is directly connected to the VCSEL diode and to a current source for actively controlling a voltage on the VCSEL diode.

[0052] A DC coupled driver for driving an optoelectronic emitter such as a VCSEL is presented, which overcomes the shortcomings of the described prior art. Having both accuracy and low power consumption, it well suited for very high speed data rates due to the incorporation of an active back termination.

[0053] Referring to FIG. 6 where a first embodiment is shown, a differential input amplifier 600 comprises a differential input npn bipolar transistor pair q₁ and q₂ with differential voltage inputs v_in and v_ip applied to each base of the transistors q₁ and q₂, respectively. The peak to peak differential signal amplitude of v_in and v_ip should be large enough to completely switch a current from one side of the differential pair to the other.

[0054] A supply voltage rail with a voltage V_dd provides electrical power to the differential input amplifier 600 through current sources I and J to each collector of the transistors q₁ and q₂, respectively. At a circuit node V_o the current source I is connected to a collector of the transistor q₁ and to an anode of a diode D, which may be a vertical cavity surface-emitting laser (VCSEL) diode, for instance. A cathode of the diode D is connected to a circuit ground.

[0055] At a circuit node V_x the current source J is connected to a collector of a transistor q₃ and to one terminal of a feedback resistor G. The other terminal of the feedback resistor G is connected to a base of the transistor q₃ and to a collector of the transistor q₂. The emitter of the transistor q₃ is connected to the anode of a diode d_g, whose cathode is connected to the circuit ground. The emitters of the transistors q₁ and q₂ are connected to the circuit ground through a tail current source 2K.

[0056] Voltage differences are sensed in a termination resistor R connected between the nodes V_o and V_x. The voltages at nodes V_o and V_x are input into a non-inverting and an inverting input of an operational amplifier (OA) U1 through resistors r₁ and r₂, respectively. A bypass capacitor c₂ connects the non-inverting input of the OA U1 to the circuit ground. The output of the OA U1 is connected to its inverting input through an integrating capacitor c₁ as well as to the input of an inverting voltage to current converter, U0. The output of the current converter U0 is drives the base of the transistor q₃.

[0057] The common mode voltage of signals v_ip and v_in should be less than two diode drops above the circuit ground reference if using the npn bipolar transistors q₁ and q₂ as shown, or less than 2*V_t above a voltage reference if using low voltage NMOS FETs m₁ and m₂ in place of the npn bipolar transistors q₁ and q₂ as shown in FIG. 7.

[0058] The input signals v_ip and v_in typically have 10 to 90% rise and fall times less than 0.35/(data rate) for best high speed performance.

[0059] An analysis of the DC bias for the differential input amplifier 600 will now be presented in the neutral input condition, i.e. when v_ip=v_in. The differential pair q₁ and q₂ splits the tail current 2K into two equal currents K such that
i₁=i₂=K  Equation [2]

[0060] A DC-coupled low frequency feedback loop comprises the OA U1, the current converter U0, and the components r₁, r₂, c₁, and c₂, which act to ensure that the current through a termination resistor R is maintained at zero. The resistors r₁ and r₂ are equal in value, which is substantially larger than the value of the termination resistor R:
r₁=r₂>>R  Equation [3]

[0061] Thus, the output voltage V_o is given by:
V_o=(I−K)*z_D  Equation [4]

[0062] where z_D is an impedance of the diode D.

[0063] Also,
V_x=V_o  Equation [5]

[0064] since no current flows in the termination resistor R. The transistor q₃ will be conducting a current (J−K), with current K flowing through the feedback resistor G. The components r₁, r₂, c₁, and c₂ along with the gain of the current converter U0 and the OA U1 set the loop bandwidth to a desired frequency, typically around 200 kHz. The current converter U0 could also be implemented with an inverting gain voltage amplifier with an output impedance much greater than the input impedance of the shunt amplifier formed by the transistor q₃ and the feedback resistor G. This latter requirement on the output impedance ensures that little of the transistor q₂ collector current flows into the current converter U0.

[0065] Note that diode d_g and the base-emitter voltage V_be of transistor q₃ in series generate a voltage drop of approximately 1.5V, which is close to a threshold voltage V_t of the VCSEL diode D. Thus a convenient bias point for modulation of the VCSEL diode D is established in a range suitable for the operation of the driver circuit 600.

[0066] Using the DC operating point for the input voltages v_ip=v_in as a reference, the static circuit currents and voltages for v_ip>v_in and for v_ip<v_in can be described as shown in Table 1:

[0067]

TABLE 1
	i1	i2	idg	iD	Vx	Vo
vip = vin	K	K	J − K	I − K	define = 0	define = 0
vip > vin	2K	0	J − 2K	I	K*G	K*zD
vip < vin	0	2K	J	I − 2K	−K*G	−K*zD

[0068] The values given in this table are accurate as long as the DC coupled low frequency feedback loop is allowed sufficient time to settle. At data rates higher than the loop bandwidth of the DC coupled low frequency feedback loop, the values of G, R, I and J need to be chosen relative to the diode impedance z_D to ensure that the average modulated voltages and currents match the DC values for the circuit.

[0069] For operation at frequencies greater than the loop bandwidth of the DC coupled low frequency feedback loop, the output voltage V_o can be expressed as the superposition of the voltage V_x multiplied by the voltage divider formed by the diode impedance z_D and the termination resistor R, plus the current flowing into node V_o multiplied by the parallel value of the termination resistor R and the diode impedance z_D. So, writing the superposition equations for the two values of v_ip and v_in, relative to the values of V_x and V_o for v_ip=V_in:
v_ip>v_in: V_o=K*G*z_D/(R+z_D)+K*R*z_D/(R+z_D)  Equation [6a]
v_ip<v_in: V_o=−K*G*z_D/(R+z_D)−K*R*z_D/(R+z_D)  Equation [6b]

[0070] Next, values for the termination resistor R and the feedback resistor G are chosen such that the DC and the high speed average modulated value of V_o are equal.

[0071] Note that:
V_o(v_ip>v_in)=−V_o(v_ip<v_in)  Equation [7]

[0072] so it is possible to write one equation for both conditions, setting the DC value equal to the modulated value:
K*z_D=K*G*z_D/(R+z_D)+K*R*z_D/(R+z_D)  Equation [8]

[0073] which simplifies to
R+z_D=G+R,  Equation [9]

[0074] which is satisfied when G=z_D.

[0075] This allows the termination resistor R to be chosen based on other considerations, such as the output impedance of the driver. If diode D is replaced by a sine wave current source operating at a frequency higher than the loop bandwidth of the DC coupled low frequency feedback loop, then it is seen that the driver output impedance value is R, since the output impedance of the shunt feedback amplifier formed by the transistor q₃ and the feedback resistor G is ideally zero.

[0076] For the best termination of a load impedance z_D,
R=z_D  Equation [10]

[0077] So, preferably, the feedback resistor G is chosen such that
G=R=z_D  Equation [11]

[0078] In practice, since the output impedance of the shunt feedback amplifier in series with the diode d_g is approximately 2/g_m, where g_m is the average transconductance of the transistor q₃ and of the diode d_g, then somewhat more practical values are:
R+2/g_m=z_D=G  Equation [12]

[0079] It has been found in practice that values larger or smaller than these calculated values still allow the driver amplifier to work well, although under modulation the current through the termination resistor R will be non-zero. What this means is that the high frequency, modulated amplitude and DC amplitude of V_o will no longer be given accurately by the above equations. Still, the circuit will swing above and below the DC operating point given by v_ip=v_in with equal amplitudes so the driver is well behaved. This is important, because impedance z_D varies somewhat from driver unit to driver unit, as do the feedback resistor G and the termination resistor R when fabricated as an integrated circuit.

[0080] From Table 1, the maximum VCSEL current, or ‘one’ level current is I. The minimum or ‘zero’ level current is I−2K. An advantageous property of this driver is that a supply current, I+J, from the voltage supply V_dd is independent of the modulation amplitude of the tail current 2K. Since the currents I and J are constant, the supply current does not change. Ideally, this means no supply bypass capacitors are needed to keep the voltage ‘quiet’ or noise-free, so that cross-talk to neighboring circuits does not result.

[0081] In practice, fast edges and slight delay mismatches in the circuit often lead to current spikes on supply voltage V_dd during edge transitions. These spikes are approximately (I+J)/10 in amplitude, with duration equal to the rise and fall time of input voltages v_ip and v_in. Since for high speed operation, the rise and fall times are very fast, perhaps 10 ps to 50 ps, good supply bypassing is best achieved with small values of capacitance having a very high self resonant frequency. Small values of capacitance are convenient for on-chip bypassing.

[0082] Choosing J=I is advantageous. If J is less than I, the collector current in the transistor q₃ may be insufficient to keep it biased with sufficient current for high speed operation. If J is greater than I, excess supply power will be dissipated.

[0083] There is an even more advantageous reason for choosing J=I. Referring to Table 1, note that if I=J, then i_dg and i_D are always equal, but opposite in phase. This means that for the driver 600 in FIG. 6, at the circuit ground node where i_dg and i_D sum together, the current is constant at (2I−2K) regardless of input voltage conditions v_ip>v_in or v_ip<v_in. Ideally then, there are no transient currents which can create ground bounce, which, if present, could cause problems within the driver and in nearby circuits.

[0084] As in the case of the supply current, the ground current at the circuit ground node also has spikes related to the input signal rise and fall time as well as to the signal delay through diode D. The phase and amplitude of spikes in the ground current and the supply current are such that placing a small capacitor between V_dd and the circuit ground node allows these currents to cancel each other, thus further improving the ‘quietness’ of the driver circuit 600.

[0085] When the diode D is a VCSEL, it is usually connected with parallel wire-bonds to the driver chip. Because of the way the currents i_dg and i_D are designed to be of equal amplitude and opposite phase, the circuit ground connection of VCSEL D will tend to extract a current from the cathode equal to the current supplied to the anode of VCSEL D, thereby ensuring a quasi-differential drive. Thus benefits can be gained in terms of speed, jitter reduction, and the confinement of electromagnetic fields emanating from the wire-bonds. This feature permits very fast operation of the driver circuit, making it suitable for operation at high data rates.

[0086] A further advantage of the driver circuit 600 is its robustness with respect to consecutive identical digit (CID) streams as long as several hundred bits. The ‘one’ and ‘zero’ levels tend not to shift very much. The exact value of CID allowed depends upon the loop bandwidth of the DC coupled low frequency feedback loop.

[0087] FIG. 7 shows the implementation of a second embodiment of the invention in deep submicron CMOS (gate length<180 nm, such as 90 nm or even shorter), where the power supply voltage V_dd may be limited typically between 1.0 V and 1.2 V. The driver 700 is similar to the driver 600 described above, however the differential input npn bipolar transistor pair q₁ and q₂ has been replaced with a differential input PMOS FET pair m₁ and m₂ having differential voltage inputs v_in and v_ip applied to their respective gate terminals, respectively. The sources of the PMOS FETs m₁ and m₂ are connected together and biased by a current source 2K from the supply voltage V_dd. An operational amplifier (OA) Ub operating from a separate supply voltage of 3 volt is added for driving the VCSEL D. The function of the termination resistor R, the diode d_fg and the PMOS FET m₃ is similar to that of the termination resistor R, the diode d_g and the transistor q₃, respectively, in the driver 600.

[0088] Since the VCSEL D requires a still higher voltage to bias it properly, OA Ub is used to supply the approximately 2.6V bias required at the anode of the VCSEL D. The non-inverting input of the OA Ub is connected to a series combination of two bias resistors R_b between the supply voltage V_dd and the circuit ground to keep the cathode of VCSEL D centered at a voltage of approximately V_dd/2 for optimum bias of the driver 700. Capacitor C_c couples the high frequency components of VCSEL current i_D back to V_dd so that the currents i_dg and i_D may still sum and cancel as described previously.

[0089] A third embodiment of the invention is shown in FIG. 8, comprising a low power DC coupled driver 800 for driving a diode D (such as a VCSEL). It is both accurate and low power, made suitable for very high speed data rates through the incorporation of an active back termination.

[0090] In particular, driver 800 exhibits stability against baseline wander during long CIDs or where the DC content of the differential input data is not balanced.

[0091] Differential voltage inputs v_ip and v_in are each connected to a gate on a differential input NMOS transistor pair m_p and m_n, respectively, each of whose drain is connected to a first supply voltage V_cc through corresponding load resistors R_cp and R_cn. Load resistors R_cp may not be necessary as in high speed circuits it tends to make the input impedance at the gate of transistor m_p equal to that at transistor m_n by providing a matched termination of capacitor C_dg of each transistor, but it is not required that this resistor be present or that it be the same value as R_cn. Sources of the transistors m_p and m_n are connected together to a current source I_m working into the ground connection. A cathode of the diode D is connected to the drain of transistor m_n, while its anode is fed from a voltage v_reg through a resistor R. The voltage v_reg in turn is generated from a second supply voltage V_dd through a current source I_b/a in series with a load resistor R*a, where ‘a’ is a constant less than 1, typically being equal to ½. A low speed feedback circuit comprising an upper current source I_b/a, transistors m_g and m_f, a feedback capacitor C_f, a feedback resistor R_f, and an OA U_f also drives the voltage v_reg. Voltages at the upper current source I_b/a and at the anode of the diode D are sensed through two resistors 2r, respectively, which are connected to a voltage node v_p at the non-inverting input of the OA U_f which is AC grounded through a bypass capacitor C to circuit ground. Voltage v_reg is sensed by resistor r connected to a voltage node v_n at the inverting input of the OA U_f. An integrating capacitor C is connected from an output of the OA U_f to the inverting input voltage node v_n.

[0092] Resistors r and 2r are much larger than resistors R and R*a. Also, the value of R is kept small relative to the impedance z_D of VCSEL diode D.

[0093] First, consider the operation of the circuit at low frequencies, where the entire feedback path is operational. A low speed feedback circuit consisting of the differential input operational amplifier U_f, the transistors m_g and m_f, the resistors R_f, r, and 2r, and both capacitors C, keeps voltages v_p and v_n equal to each other. This forces the DC and low frequency current in the resistor R to equal a current I_b, and the current in the resistor R*a to equal I_b/a. Hence, neglecting the small current through R_f (for R_f fairly large), the drain current of transistor m_f equals I_b(1/a−1).

[0094] So, the low speed feedback circuit continues to supply a constant current I_b to the diode D by adjusting the voltage v_reg to ensure that despite the modulation current impact on the current value in diode D, v_reg stays at a constant average value, which prevents baseline wander. The low speed feedback circuit acts like a current controlled voltage source, adjusting the value of v_reg to ensure that current I_b flows in diode D.

[0095] Now consider the operation of the circuit at high frequencies. Within the low speed feedback circuit just described is a high speed shunt feedback amplifier, consisting of the transistor m_f, the feedback capacitor C_f, and the feedback resistor R_f, which acts to keep the voltage v_reg constant. The goal of the high speed shunt feedback amplifier is compatible with the low speed feedback circuit, which performs the same function, but at low frequencies by ensuring a proper current split between R*a and R. As a modulation current I_m is switched into the load resistor R_cn and the impedance z_D, with I_D given by Equation 13 flowing through the diode D, this current also flows through resistor R, which causes v_reg to change.
I_D=I_m*R_cn/(R_cn+z_D+R)  Equation [13]

[0096] Since the bandwidth of m_f, C_f, and R_f is very high, this circuit immediately responds to cancel the change in current at node v_reg, thus quickly restoring v_reg to the desired value. Due to an output impedance, z_o, of the high speed shunt feedback amplifier being approximately 1/g_m of transistor m_f, there is some swing at the node v_reg as the modulation current I_m from the differential transistor pair m_p and m_n driver is switched into and out of the VCSEL diode D. The output impedance z_o is essentially added to the value of the diode impedance z_D, so that the portion of the current I_m into VCSEL diode D is split according to the equation I_m*R_cn/(R_cn+z_D+R+1/g_m). Thus the value of the current I_D into diode D is reduced by the output impedance z_o of the high speed shunt feedback amplifier, but the ability to prevent baseline wander above the low frequency corner is not impacted.

[0097] The low speed feedback circuit in concert with the high speed shunt feedback amplifier acts like a very large value of the capacitor C_g in the prior art driver 400 shown in FIG. 4. With r=5 megohm, 2r=10 megohm, C=2 pF, C_f=5 pF, R_f=2 kohm, g_m(U_f)=100 μmho, g_m(m_g)=2.5 mmho, g_m(m_f) g_m=24 mmho, and R=5 ohm, the low speed feedback circuit has a low frequency corner of about 10 kHz with a 50 degree phase margin. To achieve the 10 kHz corner frequency with the capacitor C_g in the prior art driver 400 would require
C_g=(½π*10 kHz)/(z_D+R_c)=0.11 μF,  Equation [14]

[0098] which is too large a capacitance to be practical on an integrated circuit and for which there is insufficient off-chip room in many applications.

[0099] Clearly, it is an advantage of this embodiment that the low frequency corner can be set to any desired frequency using component values compatible with on-chip integrated circuit processes in use. The penalty for this improvement in performance relative to the prior art driver 400 is an increase in power dissipation. The complete power dissipation for the low power VCSEL driver 800 given by Equation 15 constitutes an increase of V_dd*I_b*(1/a−1).
P_diss=V_cc*I_m+V_dd*I_b/a  Equation [15]

[0100] A fourth embodiment of the invention is presented in FIG. 9. A DC coupled driver 900 for driving a VCSEL with accuracy and low power, is made suitable for very high speed data rates by incorporating an active back termination.

[0101] The invention operates on the same principles as described in the first embodiment shown in FIG. 6. In the driver 900 two bipolar transistors q_1a, q_2a have been added to the transistor pair q₁, q₂, respectively, while scaling the value of several components to achieve a large reduction in power dissipation, but retaining most of the advantages of driver 600. The emitters of the four transistors q_1a, q_2a, q₁, q₂ are all connected together. However, the collector of transistor q_2a is connected to the supply voltage V_dd whereas the collectors of transistors q₁ and q_1a are connected together at node V_o to a current source I that is fed from the supply voltage V_dd.

[0102] The emitter area of the transistors q_2a is scaled to be ‘n’ times larger than for transistor q₂, so that ‘n’ times the amount of current will flow in that transistor. The factor ‘n’ is generally chosen as large as possible within the matching limitations of the devices available, a typical value would be at least 10. Thus, current i₂ is a fraction 1/(n+1) of the current source I. To maintain the voltage V_x at the same level as in the driver 600, the feedback resistor G is increased in value by the factor (n+1) to G(n+1). For similar reasons, the current source J in the driver 600 is replaced by a current source approximately equal to I/(n+1). With these changes, the driver 900 of FIG. 9 operates similar to the driver 600 of FIG. 6.

[0103] One disadvantage of the driver 900 compared to the driver 600 is that the supply current is not constant. As the modulation current 2K is switched back and forth between the differential pair transistors q_1a, q₁, and q_2a, q₂, the supply current will also be modulated. This will require more careful supply voltage bypassing than for the driver 600.

[0104] An advantage of the driver 900 over the prior art is that the current drawn from V_dd is reduced from (I+J) or 2I where I=J in the typical case, to only I(n+2)/(n+1)+K for a 50% duty cycle in the input data. In typical VCSEL driving applications, the current K tends to be approximately I/4, so for n=10, the invention requires only ⅔ as much supply current. This makes the supply current nearly the same as the very simple prior art current steering driver 100 with no back termination as shown in FIG. 1, which has power supply current I+K.

[0105] Advantageously the driver 900 also provides a controlled output impedance or back termination, that is absent in the prior art driver 100 in FIG. 1. This makes the driver 900 more suitable for operation at high data rates. Compared to the prior art driver 100, where a modulation current also appears on the supply voltage, there is no difference in bypassing requirements for the driver 900.

[0106] A fifth embodiment comprises essentially the same circuit as for driver 900, but with all the npn bipolar transistors replaced with NMOS FETs, so that the emitter, base and collector connections are substituted by source, gate and drain connections, respectively. Where the supply voltage V_dd is limited to low values such as 1.2 V or less by CMOS processes with short gate length e.g. 90 nm or shorter, the ground connection of the cathode of the diode D would need to be replaced with a connection to a negative voltage source.

[0107] In a sixth embodiment, the polarity of all the FETs is changed so that the NMOS FETS of the fifth embodiment are exchanged for PMOS FETs. This requires the circuit ground and the supply voltage connections to be swapped.

[0108] FIG. 10 shows a seventh embodiment of a VCSEL driver 1000 implemented in deep sub-micron CMOS technology (gate length less than 180 nm, such as 90 nm), in which the power supply may be limited to typically 1.0 V to 1.2 V. Along the lines of the previous embodiment 900, the driver 700 shown in FIG. 7 is modified by an expansion of the FET differential pair with two additional PMOS FETs m_1a and m_2a paralleled with the FETs m₁ and m₂ and driven by the same input gate voltages, respectively. The gate width of the PMOS FET m_2a is scaled by a factor ‘n’, analogous to scaling the emitter area of the bipolar transistor q_2a in the driver 900. The factor ‘n’ is generally chosen as large as possible within the matching limitations of the devices available, a typical value would be at least 10.

[0109] Since the VCSEL D still requires a higher voltage to bias it properly, amplifier Ub is used to supply the approximately 2.6V bias required at the anode of the VCSEL D. The midpoint of two equal series resistors R_b connected between supply voltage V_dd and circuit ground provides a voltage reference for OA Ub to keep the cathode of VCSEL D centered at approximately V_dd/2 for optimum bias of the driver 1000. Coupling capacitor C_c couples the high frequency components of diode current i_D back to supply voltage V_dd so that the currents i_dg and i_D may still sum and cancel as described previously.

[0110] An eighth embodiment of a DC coupled driver according to the invention is shown in FIG. 11. The DC coupled driver 1100 for provides an accurate and low power drive for a common anode VCSEL D is suitable for very high speed data rates due to the incorporated active back termination.

[0111] Referring to FIG. 11, a differential input voltage v_ip−v_in acting on a differential field effect transistor pair m_p and m_n switch a current i_m/n between the transistors m_p and m_n. When v_in>v_ip, the current i_n=i_m/n, and current (i_o−i_m)/n flows into a current mirror labeled c_m comprising the FETs m_n1, m_nn and m′_n1, m′_nn. The current mirror c_m provides a current gain of ‘n’, being greater than or equal to 1. In integrated circuits, this is usually achieved by making the gate width of the FET m_nn ‘n’ times that of the FET m_n1, for instance. In the case of bipolar transistors, emitter areas of the corresponding transistors are scaled. Thus the current flowing out of c_m into VCSEL diode D is (i_o−i_m). This current corresponds to a minimum drive current, commonly called a zero- or ‘0’ level current.

[0112] When v_in<v_ip, the current flowing into the current mirror c_m is i_o/n and the output of the current mirror c_m is i_o. This current corresponds to a maximum drive current, referred to as a one- or ‘1’ level drive current for the VCSEL diode D. The additional circuitry described in FIG. 11 is used to control the output impedance of the driver 1100. The value of resistor R_n is chosen to be approximately ‘n’ times the dynamic impedance of the VCSEL diode D. This means that a peak to peak voltage swing at the gate of a transistor m_t will be close to that of a peak to peak voltage swing at node v_d.

[0113] An amplifier F(s) consists of a high input impedance, low output impedance amplifier with a dominant response of a single pole low-pass filter or integrator. Negative feedback will cause the DC average value of the voltage difference v_d−v_t to be zero. If the long-term average pulse density of the differential input voltage v_ip−v_in is balanced between the ‘1’ level and the ‘0’ level, the feedback will ensure that no DC average current will flow in resistor R_t, thus keeping the voltage swing at node v_t centered around the swing at node v_d. An advantage of including resistor R_t is that it lowers the output impedance of the driver 1100.

[0114] Without the resistor R_t, the driver output impedance would be that of the current mirror c_m, which is very high. A high impedance that does not match the impedance of the VCSEL diode D will allow reflections to persist from previous data bits, which can interfere with the data bit currently transmitted, thereby increasing bit error rates and degrading data eye quality. By using the resistor R_t with the feedback circuitry that forces the DC current through the resistor R_t to zero while allowing the high speed voltage v_t to track the voltage v_d, the drive current from c_m flows only into the VCSEL diode D. Thus no drive current is wasted in resistor R_t, and the overall power dissipation is minimized.

[0115] Another advantage of this embodiment is that a delay through transistor m_t is very nearly equal to a delay of the drive current through the current mirror c_m. Thus, the voltage v_t does not lag or lead the voltage v_d significantly, thereby ensuring that the transient current through R_t is minimized, acting in unison with the feedback circuitry that minimizes the DC current flowing through R_t.

[0116] Ideally, the only current flowing through R_t is the current of a reflection it absorbs, thus preventing the re-reflection of that reflection back toward the VCSEL diode D, where it would interfere with any currently transmitted data bit. The power dissipation of the driver circuit 1100 is higher than the prior art driver 500 in FIG. 5 by approximately the factor 1+1/n.

[0117] This small increase in power dissipation is advantageous in that the circuit can be made considerably faster, because smaller currents and smaller transistors can be used to generate the driver signal prior to the current mirror c_m. By reducing the size of these transistors, an output voltage of a pre-driver circuitry used to generate signals v_ip and v_in can be scaled by the same factor of ‘n’ as well. This represents a considerable reduction in power dissipation, as roughly half the power dissipated in a typical VCSEL driver is consumed by the pre-driver circuitry.

[0118] Overall, the driver 1100 in FIG. 11 results in a reduction of total power consumption by roughly a factor of 0.5(1+2/n). For practical implementations of the driver 1100 and pre-driver circuitry, a value of n=3 results in a 20% reduction of power dissipation and a 30% reduction in driver rise and fall times.

[0119] For higher values of ‘n’, power dissipation savings would be larger, but the rise and fall time improvements decline due to gain-bandwidth limitations of current mirrors achievable in state of the art CMOS processes available for the high data rates required in state of the art drivers.

[0120] By incorporating some additional circuitry in the driver 1100 as shown in FIG. 12, a ninth embodiment of the DC coupled driver 1200 is obtained, which allows wave-shaping of the drive current.

[0121] The optical/electrical transfer function of a VCSEL D is very non-linear. One of the more important non-linearities is that the VCSEL D takes longer to turn off than to turn on. One way to balance the turn-on and turn-off times is to extract charge from the VCSEL D faster during turn-off than during turn-on. This can be accomplished by causing a falling edge of a VCSEL drive current waveform to overshoot for a duration of 20-50% of a minimum unit interval (UI) pulse width so as to reduce the turn-off time of the VCSEL D. To improve turn-on characteristics of the VCSEL optical eye diagram, it is sometimes also useful to add overshoot to a rising edge of the VCSEL drive current waveform.

[0122] A circuit which can perform these two functions starts with a delayed version of the differential input voltage v_ip and v_in. A cell marked delay adds 0.2-0.5 UI of delay to input signals v_ip and v_in and outputs them to drive a set of differential pairs comprising FETs m_f, m_v and m_r, m_c. The differential pair m_f and m_v, with current source i_fe/n provides falling edge overshoot. Similarly, the differential pair m_r and m_c, with current source i_re/n provides rising edge overshoot. The outputs of these are summed with the source and drain currents of the differential FET pair m_p and m_n to provide the desired wave-shaping, which are then amplified by the current mirror c_m. The wave-shaping requires additional supply current for operation, thus resulting in a slight increase of power dissipation. To determine the amount of power dissipation required for this circuitry, the same approach may be applied for comparing the power dissipation with the driver 1100.

[0123] The DC coupled driver 1200 represents a major improvement over the prior art common anode VCSEL driver 500 shown in FIG. 5.

[0124] A tenth embodiment comprises essentially the same circuit as for driver 1200, but with all the NMOS FETs replaced with npn bipolar transistors, so that emitter, base and collector connections are substituted for the source, gate and drain connections, respectively.