(N-1)/n current reduction scheme in a n-to-one parallel-to-serial conversion

[0002] The advent of high-speed data communications has led to new protocols being developed that transfer data at high rates of speed. Ten Gigabit Ethernet is one such protocol. Ten Gigabit Ethernet has defined two interfaces between a physical layer device (PHY) and an Ethernet controller, also referred to as a media access controller (MAC). These two interfaces are a 10 Gigabit per second (Gbps) Attachment Unit Interface, also referred to as “XAUI,” and a 10 Gbps Media Independent Interface, also referred to as “XGMII.”

[0003] XGMII is a parallel interface consisting of 32 data lines, one clock line, and four control lines in each direction. The clock runs at 156.25 megahertz and the data is clocked in and out on both the rising and falling edges of the clock. XAUI is a serial interface of four lanes of data that are 8 bit/10 bit (8 b/10 b) encoded. The encoding process takes the parallel data of XGMII and divides it into four lanes, each having 10 bits. Each lane runs at 3.125 Gigabaud in each direction. XAUI has no clock, because the clock signal is embedded in the data as part of the 8 b/10 b encoding.

[0004] XGMII is designed to be a standard parallel digital interface between the MAC and the PHY, but it is difficult to implement on standard printed circuit board (PCB) material over nominal trace distances because it is a parallel bus operating at high speeds. The alternative serial interface, XAUI, was developed to allow the distance between the PHY and the MAC to be at least 20 inches of PCB trace.

[0005] Most chips having a MAC are digital chips and therefore have a XGMII interface. Thus, devices are needed to convert XGMII to XAUI. A representative example is the Intel LXT1001 chip. The LXT1001 chip is placed near 10 Gigabit MAC chips that have a XGMII interface. The Intel LXT1001 chip accepts parallel data from the XGMII, which provides 32 data signals, four control signals and a clock signal. The data is latched into four XGMII input buffers. The data from the XGMII input buffers is encoded by an 8 b/10 b encoder. The encoding ensures that sufficient transitions are embedded to allow the receiver to recover a clock from the serial bit data stream.

[0006] After encoding, the four streams of 8 b/10 b encoded characters are each transmitted to an XAUI transmitter. The XAUI transmitter includes a pre-transmit equalizer, a parallel-to-serial converter and an XAUI output buffer. For each data stream, the data bits are clocked in parallel from the equalizer to the parallel-to-serial converter, and transmitted in serial to the output buffer.

[0007] In a high-speed serial link application, circuit designers encounter the requirement to convert a n bit parallel data stream into bits in a serial datastream. A common method to implement this conversion is utilize a n-to-1 multiplexing scheme in which n gated switches are applied to allow n bits of data go through in sequence. The outputs of the n gated switches are tied together to go to a next stage, which is most often a buffer.

[0008] When data speeds reach the Gigahertz range, a simple complimentary metal oxide semiconductor (CMOS) transmission gate switch is not sufficient. Furthermore, because of the different response speeds of positive polarity metal oxide semiconductor (PMOS) and negative polarity metal oxide semiconductor (NMOS) (NMOS is utilized to pull down the clock transition from high to low & PMOS is utilized when the transition is from low to high), the rising and falling edges of the clock signal are often different and this difference translates into clock duty cycle distortion. This problem is worsened because PMOS and NMOS vary differently over process and temperature thresholds.

[0009] FIG. 1 illustrates a current mode logic (CML) cell in use in Gigahertz circuit applications according to the prior art. In Gigahertz applications, these switches are often CML cells as shown in FIG. 1. However, the utilization of CML cells introduces a circuit with low current efficiency, meaning these circuits have a high current consumption, which is a critical parameter in many applications.

[0010] FIG. 2 illustrates a block diagram of CML cells and the passage of data through the CML cells according to the prior art. Any of these n CML cells actually pass data for only 1/N of a clock cycle and sit idle for the remaining clock cycle time ((N−1)/N). During the idle time, the tail current of the CML cell is thrown away to the power supply.

[0011] FIG. 3 illustrates a circuit diagram of a conventional multiplex implementation of parallel-to-serial conversion according to the prior art. In this embodiment, the current branches are connected to the output load one at a time. In an embodiment where the parallel-to-serial conversion is a n bit to one bit conversion, the required tail current of each of the current branches be n×I₀. In the conventional multiplex, only one of the branches is actively converting but all of the current sources are being utilized in a sequent fashion. Thus, the total current is n×n×I₀.

[0012] Accordingly, a need exists for a parallel-to-serial conversion system that reduces current consumption by a significant factor.

[0013] FIG. 1 illustrates a current mode logic cell in use in Gigahertz circuit applications according to the prior art;

[0014] FIG. 2 illustrates a block diagram of current mode logic cells and the passage of data through the CML cells according to the prior art;

[0015] FIG. 3 illustrates a circuit diagram of a conventional multiplex implementation of parallel-to-serial conversion according to the prior art;

[0016] FIG. 4 illustrates a block diagram of a conversion device according to an embodiment of the present invention;

[0017] FIG. 5 illustrates a circuit diagram for a first conversion module for a conversion device according to an embodiment of the present invention; and

[0018] FIG. 6 illustrates a circuit diagram for a second conversion module for a conversion device according to an embodiment of the present invention.

[0019] The present invention converts a parallel datastream into a serial bit datastream in a current efficient manner. FIG. 4 illustrates a block diagram of a conversion device according to an embodiment of the present invention. The conversion device results in a reduction in current consumption by a factor of (n−1)/n during a n-to-1 parallel-to-serial conversion. The conversion device includes a plurality of current sources 10-16, a plurality of switching devices 18-24, a current steering device 26, and a voltage output device 28.

[0020] The voltage output device 28 provides an output voltage representative of the value of a bit in the serial datastream that has been converted from a parallel bit in the parallel datastream. The value of the serial bit is the current going through the resistor, e.g., n×I_o, multiplied by the resistance/impedance, e.g., n×I_o×R, of the voltage output device 28. In the present invention, the current going through resistor is the combined current with the magnitude of the number of bits being converted from parallel to serial, e.g., n, times the magnitude of the current generated from each current source.

[0021] Depending on the how the circuit is designed, the magnitude of the voltage amplitude required in order to identify if a signal is, illustratively, high or low, may vary. As discussed previously, the output voltage is equal to I_o×R. In a high-speed circuit, including circuits operating in the Gigahertz range, the resistance, R, has an upper limit. The upper limit of resistance, illustratively R, is a function of the highest signal frequency and the capacitance load. If the loading capacitance for the parallel-to-serial conversion device is C_o, R cannot be greater than R=1/(2πf_cC_o), where f_cis the highest frequency to which the circuit is able to respond.

[0022] Therefore, if the specification for the input voltage requires a signal amplitude of V_oand the maximum resistance is R, then at least V_o/R current is used to meet the specification. Let the required current, V₀/R, equal n×I₀. In the conventional design, each current branch, or tail current for each bit, has a required current of the magnitude, e.g., n×I₀. However, with an embodiment of the present invention, each current branch, or tail current for each bit, has a magnitude of 1/nth the required current, e.g., I₀. Thus, the total branch current (including n branches) for the n bit conversion is n×n×I₀in the conventional design and n×I₀for an embodiment of the present invention. Therefore, the reduction in current compared to the common multiplex is (N−1/N).

[0023] The number of current sources 10-16 and selection devices 18-24 are directly proportional to the number of parallel bits being converted into a serial bit stream. Illustratively, an n-to-1 bit parallel-to-serial conversion requires that n current sources are present in the conversion device. For example, in an 8-to-1 parallel-to-serial conversion, eight current sources are present in the conversion device. Similarly, in a 10-to-1 parallel-to-serial conversion, ten selection devices are present in the conversion device.

[0024] As illustrated in FIG. 4, the current flowing through voltage output device 28 reaches point B 11 in the diagram and may be split into a first current C113 and a second current C215. Because only one bit is being converted at one time, only one selection device is active at each time. Thus, one selection device is an active selection device and (n−1) selection devices are non-active selection devices.

[0025] If the selection device 18-24 is active (meaning the bit, D_i, input into the selection device 18-24 is being converted from a parallel bit to a bit in a serial datastream), an active selective device (one of selection devices 18-24) accepts as input the first current C113 and connects the first current directly to the corresponding current source. This current may be referred to as the active selection device current. The current may have a magnitude of Io.

[0026] For example, as illustrated in FIG. 4, if the second bit is currently undergoing a parallel-to-serial conversion, then selection device 20 is directly coupled to current source 12 and selection device current 21 is the active selection device current.

[0027] If the selection device is inactive (meaning the bit, D_i, input into the selection device is not being converted), the non-active selection devices do not draw current. Therefore, the remaining current from the voltage output device becomes the second current C215 and is directed towards the current steering device 26. Because (n−1) selection devices are inactive at all times when a parallel-to-serial conversion is taking place, the current steering device 26 may receive a current with a magnitude of (n−1)×Io. The current steering device 26 directs the second current to the current sources 10-16 which correspond to the non-active selection devices. The currents being directed to the current sources 10-16 from the current steering device 26 may be referred to as the non-active selection device currents.

[0028] Utilizing the same example as above, selection device 20 is the active selection device. Thus, selection device 18, selection device 22, and selection device 24 are inactive. Therefore, no current flows through current paths I₁39, I₃41, and I₄42. Instead, the second current C215 flows to the current steering device 26. The current steering device 26 receives the second current C215 as an input and outputs three non-active selection device currents, e.g., SI₁47, SI₃49, and SI₄50, to the corresponding current sources 10, 14, and 16, respectively.

[0029] A plurality of clock signals, e.g., ck₁, ck₂, ck₃, and ck₄, may be input into the selection devices 18-24 along with the parallel data bits, e.g., D₁-D₄. The plurality of clock signals identifies which selection device is active, e.g., currently converting the parallel bit. Only one of the parallel bits may be converted at each moment in time, so only one clock signal may indicate that its selection device is active. The clock signal may indicate a selection device is active by, illustratively, going from a low state to a high state. Conversely, the clock signal may indicate the selection device is inactive by, illustratively, going from a high state to a low state.

[0030] For example, as illustrated in FIG. 4, if the fourth parallel bit D₄is currently being converted to a bit in a serial datastream, then clock signal ck₄indicates that selection device 24 is active. Therefore, current I₄42 is switched directly to current source 16 and selection device current 25 is the active selection device current. Clock signals ck₁, ck₂, and ck₃do not undergo a transition from low to high and therefore switching devices 18, 20, and 22 are inactive, and no current flows through switching devices 18, 20, and 22. The current steering device 26 steers the second current C215 to the corresponding current sources, e.g, 10, 12, and 14. This configuration allows the use of only n×I_ocurrent for all n tail current sources, which as mentioned previously is a significant reduction from the use in the prior art of (n×n×I_o) current.

[0031] FIGS. 5 and 6 illustrate a specific circuit diagram of an embodiment of the present invention. FIG. 5 illustrates a first conversion module for a first bit in a n-to-1 bit parallel-to-serial conversion. FIG. 6 illustrates a second conversion module for remaining (n−1) bits in a parallel-to-serial conversion. For example, if a conversion device requires an eight-to-one bit parallel-to-serial conversion, one first conversion module is required along with seven second conversion modules.

[0032] A first conversion module may include transistors MN450, MN552, MN654, M1456, MN858, MN960, MN1062, and MN1164, resistors RD166 and RD268, and a voltage supply VDD 70. In an embodiment of the present invention, the transistors may be NPN transistors. Illustratively, the first terminals of the transistors are the collector; the second terminals of the transistors are the base; and the third terminals of the transistors are the emitters. Alternatively, the transistors may be field-effect transistors. Illustratively, the first terminals of the transistors are the drain; the second terminals of the transistors are the gate, and third terminals of the transistors are the source. In addition, there is a fourth terminal for field-effect transistor, which is the body.

[0033] A first terminal of MN450 may be connected to a 1st end of resistor RD166, a 1st terminal of MN552, and a 2nd terminal of MN654. A measuring point node DON 72 may be located at the node of the 1st terminal of MN450, the first end of RD166, the 1st terminal of MN552, and the 2nd terminal of MN654. A second end of resistor RD166 may be connected to the voltage supply VDD 70.

[0034] A 1st terminal of M1456 may be connected to a first end of resistor RD268, a 2nd terminal of MN552, and a 1st terminal of MN654. A measuring point node DOP 71 may be located at the node of the 1st terminal of M1456, the first end of RD268, the 2nd terminal of MN552, and the 1st terminal of MN654. A second end of resistor RD268 may be connected to the voltage supply VDD 70. A third terminal of M1456 and a third terminal of MN450 are connected to a 1st terminal of MN858. In addition, if the invention is implemented utilizing field-effect transistors, a 4th terminal of MN450, M1456, MN552, MN654, MN858, MN960, MN1062, and MN1164 may be all connected together to a reference voltage, which may be a ground.

[0035] In the first conversion module, as illustrated in FIG. 5, a 3rd terminal of MN552 and MN654 may be connected to a 1st terminal of MN960. A 3rd terminal of MN858 and MN960 may be connected to a 1st terminal of MN1062. A 3rd terminal of MN1062 may be connected to the 1st terminal of MN1164 in a cascade configuration.

[0036] Referring to FIG. 6, each second conversion mechanism may include transistors MN476, M1478, MN880, MN982, MN1084, and MN1186. The node DON 95 may be connected to a 1st terminal of MN476. The node DOP 97 may be connected to a 1st terminal of M1478. DON 95 and DOP 97 may also be connected to DON 72 and DOP 71 (see FIG. 5), respectively. That means the resistors in FIG. 5 are shared for all the modules. A third terminal of MN476 and M1478 may both be connected to a 1st terminal of MN880. Nodes Iout 110 may all be connected to a 1st terminal of MN982. A third terminal of MN880 and MN982 may be connected to a 1st terminal of MN1084. A third terminal of MN1084 may be connected to a 1st terminal of MN1186 in a cascade configuration. In addition, if the invention is implemented utilizing field effect transistors, the 4th terminals of MN476, M1478, MN880, MN982, MN1084, and MN1186 may all be connected together to a reference voltage, which may be a ground.

[0037] The conversion of a parallel bit to a bit in a serial datastream occurs in the upper section of the first and second conversion modules illustrated in FIGS. 5 and 6, respectively. For example, as illustrated in FIG. 5 for a first conversion module, a parallel bit, which may be represented as a differential signal D_i90 and D_ib92, is input into the 2nd terminal of transistors MN450 and M1456, respectively. If the parallel bit is, illustratively, high to represent a 1 value, then D_i90 is a higher value than D_ib92. Because the parallel bit is, illustratively, high, transistor MN450 is switched on and current flows through resistor RD166. The corresponding node voltage at DON 72 is VDD 70 minus the product of the current value, e.g., n×I_o, times the value of the resistance, e.g., RD166. In other words, node voltage at DON 72=Vdd−(n×I₀×RD1).

[0038] In the example, no current flows through the transistor M1456 because it is not switched on. Therefore, the value of DOP 71 is equal to Vdd 70 (Vdd=DOP). The output voltage for the circuit is defined as DOP 71−DON 72. In this case, where D_i90 is higher than D_ib92, DOP 71 is greater than DON 72 (Vdd compared to (Vdd−(n×I_o×RD1))), and the output voltage is, illustratively, high. The output signal magnitude is (Vdd−(Vdd−n×I_o×RD1)), which is equal to I_o×RD1.

[0039] In contrast, if D_ib92 is higher than D_i90 in the first conversion module, then the parallel bit has, illustratively, a low value, e.g., 0. Transistor M1456 is switched on and the current flows through resistor RD268. In this case, the voltage at node DON 72 is Vdd 70, and the voltage at node DOP 71 is (Vdd−(n×I_o×RD1)). Therefore, the output voltage is, illustratively, low, and the output signal magnitude is equal to (−n×I_o×RD2).

[0040] Referring to FIG. 6, if a second conversion module is active, then either MN476 or M1478 may be switched on, depending on whether D_i91 is higher than D_ib93 (a high condition turning MN476 on) or D_ib93 is higher than D_i91 (a low condition turning M1478 on). This setup establishes a current path from the current source, e.g., MN1084 and MN1186, through MN880 to MN476 if a high condition is present or through MN880 to M1478 if a low condition is present. The measurement point nodes DON 95 and DOP 97 for all of the second conversion modules are tied together with the measuring point nodes DON 72 and DOP 71 (see FIG. 5) of the first conversion module. Because of this configuration and measuring point nodes DON 95 and DOP 97 being connected to the first terminals of RD166 and RD268, respectively, the current flowing through RD166 and RD268 is directed to the active second conversion module. Therefore, the voltage produced across RD166 and RD268 (see FIG. 5) is representative of the value of the active second conversion module's parallel data bit using the same analysis discussed previously and the output signal magnitude is n×Io multiplied by either RD1 (if the value of the parallel data bit is high) or RD2 (if the value of the parallel data bit is low). Usually, RD1 and RD2 have equal values.

[0041] The output voltage Vo necessary to indicate the high value of a serial bit in the datastream is dependent on the specification of the device receiving the datastream. Depending on how the receiving circuit is designed, the magnitude of the voltage amplitude required to identify if a signal is, illustratively, high or low may vary. As discussed previously, the output voltage is equal to n×Io×R. In a high-speed circuit, including circuits operating in the Gigahertz range, the resistance, R, has an upper limit. As identified previously, R cannot be greater than R_crt=1/(2πf_cC₀), where C is the loading capacitance of the conversion mechanism.

[0042] Therefore, if a specification requires a signal amplitude of V_oand the maximum resistance is illustratively R, then at least V_o/R current is used to meet the specification. Therefore, the tail current for each selection device (or conversion device) needs to be at least (V_o/R_crt/n). Note this is different from a conventional design, where Vo/Rcrt is necessary.

[0043] The tail current magnitude is generated for the first and second conversion modules in the bottom section of FIGS. 5 and 6. In both the first and second conversion modules, transistors MN10 and MN11 (for the first conversion mechanism MN1064 and MN1166 in FIG. 5; for the second conversion modules MN1084 and MN1186 in FIG. 6) provide the tail current I₀specified by the application of a input voltage bias VB396 to a 2nd terminal of MN1062 or 84 and the application of an input voltage bias VB294 to a 2nd terminal of MN1164 or 86. If the specification calls for a specific current level, bias VB294, VB396 and physical device sizes of MN10, MN11 may be adjusted to obtain the required/requested level.

[0044] As illustrated in FIGS. 5 and 6, a plurality of differential clock signals, such as ck1100, ck1b102, ckn 104, and cknb 106, may be utilized to indicate which conversion module is actively converting the parallel bit to a bit in the serial data stream. For example, if the seventh bit is currently being converted, the sixth second conversion module may be currently converting a parallel bit to a bit in a serial datastream, and the differential clock signal may be, illustratively, high to indicate that a parallel bit in that module is being converted. When the seventh bit is being converted, if ck₇is higher than ck_7b, then the clock signal is high for the sixth second conversion module and the data bit D_iinput into that second conversion module is actively being converted from the parallel bit into the bit in the serial datastream. Only one parallel bit is being converted at a single moment in time, so only one clock signal may be, illustratively, high at a single time.

[0045] If the first conversion module is active, the current is steered directly through that selection module to the current source. The current is steered by the same control signal, D_i90 or D_i91, with the corresponding clock signal, ck₁, is high. In the embodiment of the invention illustrated in FIG. 5, if ck₁100 is greater than ck_1b102, then ck₁100 is high and the first conversion module is actively converting a parallel bit into a bit in a serial datastream, so MN858 is switched on. In relation to FIG. 4, transistors MN450 and M1456 and MN858 may all comprise a selection device. When MN858 is activated, a current path is established so current flows through MN450 or M1456, plus MN858, MN1062, and MN1164. The value of the parallel bit, D_i90 or D_ib92, determines whether current flows through MN450 or M1456, i.e., if D_i90 is greater than D_ib92, current flows, through MN450, and if D_ib92 is greater than D_i90, current flows through M1456.

[0046] If the first conversion module is not active, then current is steered first to a current steering device and then to the current source. If the first conversion module is inactive, e.g., ck_1b102 is greater than ck₁100, then ck₁100 is, illustratively, low, and transistor MN960 is activated. Because MN450, M1456, and MN858 are not activated, current, from the resistor RD1 or RD2, depending on whether the bit being convert currently is high or low, flows through the cross-coupled pair MN552 and MN654 to MN960. The cross-coupled pair MN552 and MN654 in combination with MN960 represent the current steering device illustrated in FIG. 4. MN450 and M1456 receive no current. In other words, since a different bit other than the 1st bit is being converted at that time, nodes DOP/DON (71/72 in FIG. 5 and 97/95 in FIG. 6) will be controlled by the bit that is being converted at the moment. Because the cross-coupled pair MN5/MN6 is designed to be relative weaker than input devices MN4/MN14, the cross-coupled pair will subdue to the active bit polarity. In the case that there is no bit being transferred, then DOP 71 and DON 72 stay with the last bit value, and MN552 and MN654 may act as a latch.

[0047] In addition, because the cross-coupled pair provides an extra current path, the current, n×I₀, going through resistors is much larger than the current, I₀, going through devices MN450, M1456, MN859, and MN960. In another words, the cross-coupled pair allows large signal swing at output, DOP/DON, without actually running large currents in these switching devices MN450, M1456, MN859, and MN960. Because the current flowing through MN450, M1456, MN859, and MN960 is small, the physical size of these transistors may be small. The small size of MN450, M1456, MN859, and MN960 reduce the loading for the circuits that drive these devices and lead to a reduction in current consumption in the circuits. This is important in high frequency design because too much loading results in high current consumption.

[0048] In all of the second conversion modules that are inactive, as illustrated in FIG. 6, ck_nb106 is greater than ck_n104 and MN982 is switched on. The node I_out110 is connected to I_in99 (see FIG. 5), and the 1st terminal of MN982. The 1st terminals on all the second conversion module's MN9 transistors 82 are connected to the third terminals of MN552 and MN654 in the first conversion module, as illustrated in FIG. 5. The coupling of all the inactive second conversion modules MN982 transistors to the third terminals of MN552 and MN654 provides the current path to resistors RD1 and RD268 for all of the inactive conversion module's respective current sources. For all of the second conversion modules that are inactive, a current path is established from either MN552 or MN654, depending on whether Di 91 is high or not, through MN982 to the current source, which are transistors MN1084 and MN1186.

[0049] The present invention illustrates a n-to-1 parallel-to-serial conversion device that includes a voltage output device, n selection devices, a current steering device, and n current sources. N parallel data bits are input into n selection devices. An active selection device is coupled directly to a corresponding current source and provides a bit to the voltage output device. The magnitude of the current from the current source is a predetermined value. The voltage device outputs a value representative of the bit in a serial datastream. The n−1 inactive selection devices do not draw current and a second current is directed to a current steering device. The magnitude of the current in the second current branch is (n−1) times the value of the active selection device current. The current steering device distributes the second current to the (n−1) current sources that correspond to the (n−1) inactive selection device currents. The current steering device allows a path to be established in order that all current is utilized within the parallel-to-serial conversion device.

[0050] While the description above refers to particular embodiments of the present invention, it will be understood that many modifications may be made without departing from the spirit thereof. The accompanying claims are intended to cover such modifications as would fall within the true scope and spirit of the present invention. The presently disclosed embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims, rather than the foregoing description, and all changes that come within the meaning and range of equivalency of the claims are intended to be embraced therein.