Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
ORTHOGONAL DIFFERENTIAL VECTOR SIGNALING CODES WITH EMBEDDED CLOCK
Document Type and Number:
WIPO Patent Application WO/2017/091715
Kind Code:
A1
Abstract:
Orthogonal differential vector signaling codes are described which support encoded subchannels allowing transport of distinct data and clocking signals over the same transport medium. Embodiments are described which are suitable for implementation in both conventional high-speed CMOS and DRAM integrated circuit processes.

Inventors:
SHOKROLLAHI AMIN (CH)
Application Number:
PCT/US2016/063575
Publication Date:
June 01, 2017
Filing Date:
November 23, 2016
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
KANDOU LABS SA (CH)
INVENT MINE LLC (US)
International Classes:
H04L25/02; H04L25/49
Foreign References:
US20150222458A12015-08-06
US20150249559A12015-09-03
US20110268225A12011-11-03
US20110268225A12011-11-03
US201113030027A2011-02-17
US201414158452A2014-01-17
US201313842740A2013-03-15
US201514816896A2015-08-03
US9100232B12015-08-04
US201461934807P2014-02-02
US201361839360P2013-06-25
US201461946574P2014-02-28
US201514711528A2015-05-13
US201462015172P2014-06-20
US201313895206A2013-05-15
US201414315306A2014-06-25
Attorney, Agent or Firm:
IRVINE, Robert J. (Suite 102Chicago, Illinois, US)
Download PDF:
Claims:
CLAIMS

We claim:

1. An apparatus comprising:

a first subchannel encoder configured to receive a first input signal and to generate elements of a first weighted subchannel vector;

a second subchannel encoder configured to receive a second input signal, and to generate elements of a second weighted subchannel vector, wherein the second weighted subchannel vector is mutually orthogonal to the first weighted subchannel vector and wherein the second input signal is asynchronous with respect to the first input signal;

an asynchronous codeword generator connected to the first and second subchannel encoders, the asynchronous codeword generator configured to receive the first and second subchannel vectors and to responsively generate elements of an asynchronous-transmit codeword by adding the first and second weighted subchannel vectors, wherein elements of the asynchronous-transmit codeword transition asynchronously in response to transitions of elements of the first or second weighted subchannel vector; and

a driver configured to transmit the elements of the asynchronous-transmit codeword as analog signals on a multi-wire bus.

2. The apparatus of claim 1, wherein the first and second subchannel encoders are analog encoders, and wherein the asynchronous codeword generator comprises one or more analog combination units.

3. The apparatus of claim 1, wherein the first and second subchannel encoders are digital encoders configured to generate a pair of sets of subchannel bits representing the first and second weighed subchannel vectors.

4. The apparatus of claim 3, wherein the asynchronous codeword generator comprises a plurality of digital adders, the plurality of digital adders configured to receive the pair of sets of subchannel bits and to responsively generate a set of codeword bits representing the asynchronous-transmit codeword; and, wherein the driver comprises a signal-level conversion circuit connected to the plurality of digital adders, the signal-level conversion circuit configured to generate the analog signals on the multi-wire bus based on the set of codeword bits.

5. The apparatus according to claims 3 or 4, wherein the asynchronous codeword generator comprises a plurality of sets of driver circuits, each set of the plurality of sets of driver circuits connected to a respective wire of the multi-wire bus, and each set of driver circuits configured to receive respective subsets of the pair of sets of subchannel bits, the respective subsets corresponding to an element of the asynchronous-transmit codeword for generation of an analog signal representing the corresponding asynchronous-transmit codeword element for transmission on the respective wire of the multi-wire bus.

6. The apparatus as in one of claims 1 -5, wherein the second input signal has a ½ unit interval (UI) phase offset with respect to the first input signal.

7. The apparatus as in one of claims 1 -6, wherein the first input signal is a data signal and the second input is a clock signal.

8. The apparatus as in one of claims 1 -7, wherein the first input signal transitions at a first rate, and the second input signal transitions at a second rate that is an integer fraction of the first rate.

9. A method comprising:

receiving, at a first subchannel encoder, a first input signal and responsively generating elements of a first weighted subchannel vector;

receiving, at a second subchannel encoder, a second input signal and responsively generating elements of a second weighted subchannel vector, wherein the second weighted subchannel vector is mutually orthogonal to the first weighted subchannel vector and wherein the second input signal is asynchronous with respect to the first input signal;

generating, using an asynchronous codeword generator, elements of an asynchronous-transmit codeword by adding the first and second weighted subchannel vectors, wherein elements of the asynchronous-transmit codeword transition asynchronously in response to transitions of elements of the first or second weighted subchannel vector; and

transmitting the elements of the asynchronous-transmit codeword as multilevel analog signals on a multi-wire bus.

10. The method of claim 9, wherein generating the first and second subchannel vectors comprises analog computation, and wherein generating the asynchronous codeword comprises analog signal combination.

1 1. The method of claim 9, wherein generating the first and second subchannel vectors comprises digital encoding configured to represent each element of the first and second weighed subchannel vectors as a set of one or more bits.

12. The method of claim 1 1, wherein the elements of the asynchronous-transmit codeword are sets of codeword bits that are generated using digital adders based on the sets of one or more bits representing the elements of the first and second weighted subchannel vectors; and

wherein transmitting the elements of the asynchronous-transmit codeword comprises converting the sets of codeword bits into analog signal levels for transmission over the multi-wire bus.

13. The method according to claims 11 or 12, wherein each set of one or more bits is generated based on logical combinations of the first and second input signals with bits representing first and second subchannel vectors, respectively.

14. The method as in one of claims 9-13, wherein the second input signal has a ½ unit interval (UI) phase offset with respect to the first input signal.

15. The method as in one of claims 9-14, wherein the first input signal is a data signal and the second input signal is a clock signal.

Description:
ORTHOGONAL DIFFERENTIAL VECTOR SIGNALING CODES

WITH EMBEDDED CLOCK

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to U.S. Application No. 14/952,492, filed November 25, 2015, entitled "ORTHOGONAL DIFFERENTIAL VECTOR

SIGNALING CODES WITH EMBEDDED CLOCK", reference of which is hereby incorporated in its entirety.

REFERENCES

[0002] The following references are herein incorporated by reference in their entirety for all purposes:

[0003] U.S. Patent Publication No. 2011/0268225 of U.S. Patent Application No. 12/784,414, filed May 20, 2010, naming Harm Cronie and Amin Shokrollahi, entitled "Orthogonal Differential Vector Signaling", hereinafter identified as [Cronie I];

[0004] U.S. Patent Application No. 13/030,027, filed February 17, 2011, naming Harm Cronie, Amin Shokrollahi and Armin Taj alii, entitled "Methods and Systems for Noise Resilient, Pin-Efficient and Low Power Communications with Sparse Signaling Codes", hereinafter identified as [Cronie II];

[0005] U.S. Patent Application No. 14/158,452, filed January 17, 2014, naming John Fox , Brian Holden, Peter Hunt, John D Keay, Amin Shokrollahi, Richard Simpson, Anant Singh, Andrew Kevin John Stewart, and Giuseppe Surace, entitled "Chip-to- Chip Communication with Reduced SSO Noise", hereinafter identified as [Fox I];

[0006] U.S. Patent Application No. 13/842,740, filed March 15, 2013, naming Brian Holden, Amin Shokrollahi and Anant Singh, entitled "Methods and Systems for Skew Tolerance in and Advanced Detectors for Vector Signaling Codes for Chip-to-Chip Communication", hereinafter identified as [Holden I]; [0007] US Patent application 14/816,896 filed August 3, 2015, naming Brian Holden and Amin Shokrollahi, entitled "Orthogonal Differential Vector Signaling Codes with Embedded Clock", naming Brian Holden and Amin Shokrollahi, entitled "Orthogonal Differential Vector Signaling Codes with Embedded Clock", hereinafter identified as [Holden II];

[0008] U.S. Patent No. 9,100,232 corresponding to U.S. Patent Application

14/612,241, filed February 2, 2015 and issued August 4, 2015, naming Ali Hormati, Amin Shokrollahi, and Roger Ulrich, entitled "Methods and Apparatus for Low Power Chip-to-Chip Communications with Constrained ISI Ratio", hereinafter identified as [Hormati I];

[0009] U.S. Provisional Patent Application No. 61/934,807, filed February 2, 2014, naming Amin Shokrollahi, entitled "Vector Signaling Codes with High pin-efficiency and their Application to Chip-to-Chip Communications and Storage", hereinafter identified as [Shokrollahi I];

[0010] U.S. Provisional Patent Application No. 61/839,360, filed June 23, 2013, naming Amin Shokrollahi, entitled "Vector Signaling Codes with Reduced Receiver Complexity", hereinafter identified as [Shokrollahi II];

[0011] U.S. Provisional Patent Application No. 61/946,574, filed February 28, 2014, naming Amin Shokrollahi, Brian Holden, and Richard Simpson, entitled "Clock Embedded Vector Signaling Codes", hereinafter identified as [Shokrollahi III];

[0012] U.S. Patent Application No. 14/711,528, filed May 13, 2015, naming Amin Shokrollahi, entitled "Vector Signaling Codes with Improved Noise Margin", hereinafter identified as [Shokrollahi IV];

[0013] U.S. Provisional Patent Application No. 62/015,172, filed July 10, 2014, naming Amin Shokrollahi and Roger Ulrich, entitled "Vector Signaling Codes with Increased Signal to Noise Characteristics", hereinafter identified as [Shokrollahi V];

[0014] U.S. Patent Application No. 13/895,206, filed May 15, 2013, naming Roger Ulrich and Peter Hunt, entitled "Circuits for Efficient Detection of Vector Signaling Codes for Chip-to-Chip Communications using Sums of Differences", hereinafter identified as [Ulrich I];

[0015] U.S. Patent Application No. 14/315,306, filed June 25, 2014, naming Roger Ulrich, entitled "Multilevel Driver for High Speed Chip-to-Chip Communications", hereinafter identified as [Ulrich II].

FIELD OF THE TECHNOLOGY

[0016] Described embodiments relate generally to the field of communications, and more particularly to the transmission and reception of signals capable of conveying information within and between integrated circuit devices.

BACKGROUND

[0017] In communication systems, a goal is to transport information from one physical location to another. It is typically desirable that the transport of this information is reliable, is fast and consumes a minimal amount of resources. One common information transfer medium is the serial communications link, which may be based on a single wire circuit relative to ground or other common reference, or multiple such circuits relative to ground or other common reference. A common example uses singled-ended signaling ("SES"). SES operates by sending a signal on one wire, and measuring the signal relative to a fixed reference at the receiver. A serial communication link may also be based on multiple circuits used in relation to each other. A common example of the latter uses differential signaling ("DS").

Differential signaling operates by sending a signal on one wire and the opposite of that signal on a matching wire. The signal information is represented by the difference between the wires, rather than their absolute values relative to ground or other fixed reference.

[0018] There are a number of signaling methods that maintain the desirable properties of DS while increasing pin efficiency over DS. Vector signaling is a method of signaling. With vector signaling, a plurality of signals on a plurality of wires is considered collectively although each of the plurality of signals might be independent. Each of the collective signals is referred to as a component and the number of plurality of wires is referred to as the "length" of the vector. In some embodiments, the signal on one wire is entirely dependent on the signal on another wire, as is the case with DS pairs, so in some cases the length of the vector might refer to the number of degrees of freedom of signals on the plurality of wires instead of exactly the number of wires in the plurality of wires.

[0019] Any suitable subset of a vector signaling code denotes a "subcode" of that code. Such a subcode may itself be a vector signaling code. With binary vector signaling, each component or "symbol" of the vector takes on one of two possible values. With non-binary vector signaling, each symbol has a value that is a selection from a set of more than two possible values. When transmitted as physical signals on a communications medium, symbols may be represented by particular physical values appropriate to that medium; as examples, in one embodiment a voltage of 150 mV may represent a symbol and a voltage of 50 mV may represent a symbol, while in another embodiment may be represented by 800 mV and as -800 mV.

[0020] A vector signaling code, as described herein, is a collection C of vectors of the same length N, called codewords. The ratio between the binary logarithm of the size of C and the length N is called the pin-efficiency of the vector signaling code. The Orthogonal Differential Vector Signaling (ODVS) codes of [Cronie I], [Cronie II], [Fox I], [Shokrollahi I], [Shokrollahi II], and [Shokrollahi III] are examples of vector signaling codes, and are used herein for descriptive purposes.

[0021] FIG. 1 illustrates a prior art communication system employing vector signaling codes. Bits So, Si, S2 enter block-wise 100 into an encoder 105. The size of the block may vary and depends on the parameters of the vector signaling code. The encoder generates a codeword of the vector signaling code for which the system is designed. In operation, the encoder may generate information used to control PMOS and NMOS transistors within driver 110, generating voltages or currents on the N communication wires 125 comprising the communications channel 120. Receiver 132 reads the signals on the wires, possibly including amplification, frequency compensation, and common mode signal cancellation. Receiver 132 provides its results to decoder 138, which recreates the input bits at 140, here shown as received bits Ro, Ri, R 2 . [0022] Depending on which vector signaling code is used, there may be no decoder, or no encoder, or neither a decoder nor an encoder. For example, for the 8b8w code disclosed in [Cronie II], both encoder 112 and decoder 1138 exist. On the other hand, for the Hadamard code disclosed in [Cronie I], an explicit decoder may be unnecessary, as the system may be configured such that receiver 132 generates output bits 140 directly.

[0023] The operation of the transmitting device 110, comprising input data 100 and elements 112 and 118, and that of the receiving device 130, consisting of element 132, optional element 138, and output data 140, have to be completely synchronized in order to guarantee correct functioning of the communication system, accurately capturing received signals from each wire 125 and presenting the received results as complete codewords to decoder 138 for analysis. In some embodiments, this synchronization is performed by an external clock shared between the transmitter and the receiver. Other embodiments may combine the clock function with one or more of the data channels, as in the well-known Biphase encoding used for serial

communications.

[0024] One important example is provided by memory interfaces in which a clock is generated on the controller and shared with the memory device. The memory device may use the clock information for its internal memory operations, as well as for I/O. Because of the burstiness and the asynchronicity of memory operations, the I/O may not be active all the time. Moreover, the main clock and the data lines may not be aligned due to skew. In such cases, additional strobe signals are used to indicate when to read and write the data.

BRIEF DESCRIPTION

[0025] Orthogonal differential vector signaling codes providing transport for both data and a clocking signal are described which are suitable for implementation in both conventional high-speed CMOS and DRAM integrated circuit processes. Example channels derived from current practice for Low-Powered DDR4 interfaces are described, as are modest channel enhancements providing higher speed and greater signal integrity. BRIEF DESCRIPTION OF FIGURES

[0026] FIG. 1 illustrates a prior art communication system employing vector signaling codes.

[0027] FIG. 2 illustrates one embodiment of an ODVS communications system in which a discrete decoding function is not required.

[0028] FIG. 3 is a block diagram of an embodiment transporting data and a clock signal using an ODVS code wherein the signal transitions of the clock being transported are offset from signal transitions of the data being transported.

[0029] FIG. 4 is a block diagram of one embodiment of an ENRZ transmitter encoding the set of data inputs as a codeword, the elements of which are represented as signaling levels and then transmitted.

[0030] FIG. 5 is a block diagram of a second embodiment of an ENRZ transmitter encoding individual data inputs on three subchannels that are then summed, represented as signaling levels, and transmitted.

[0031] FIG. 6 is a block diagram of a third embodiment of an ENRZ transmitter encoding two data inputs as codewords representing a summation of two subchannels, encoding a third data input on a third subchannel, all subchannels then being summed, represented as signaling levels, and transmitted.

[0032] FIG. 7 illustrates an Ulrich-style driver configured to combine a subchannel- encoded clock signal and codeword-encoded data onto the same wire.

[0033] FIG. 8 illustrates an Ulrich-style driver configured to combine codeword- encoded and subchannel-encoded signals from different clock domains onto the same wire.

[0034] FIG. 9 illustrates an embodiment of a receiver incorporating multi-phase processing.

[0035] FIGs. 10A and 10B show embodiments of clock phase adjustment. [0036] FIG. 11 is a flow chart illustrating the method of producing physical output signals encoding a first and a second input signal.

DETAILED DESCRIPTION

[0037] FIG. 2 is a block diagram of a communications system using a vector signaling code to transport three data bits 100 over a communications channel 120 comprised of four wires 125. For descriptive convenience and without implying limitation, a vector signaling code derived from the order-4 Hadamard matrix commonly called "H4" and "ENRZ" is used in this and subsequent examples.

[0038] ENRZ may be detected efficiently using multi-input comparators, as described in [Holden I] and subsequently described herein in greater detail. Receiver 130 incorporates three instances of multi-input comparators 233 to obtain three results derived from combinatorial manipulation of sets of received wire signals, the results of which are then measured by digital comparators 234 to obtain three received data bits 140. The receiver may optionally include amplification and/or frequency compensation (Continuous Time Linear Equalization or CTLE, as one example) 231 to compensate for attenuation or frequency-dependent loss in communications channel 120.

System Environment

[0039] One important application for a high speed chip-to-chip communications interface is to provide connection between a memory controller and one or more memory storage devices. In such applications, a clock is generated on the controller and shared with the memory device or devices. The memory device may use the clock information for its internal memory operations, as well as for I/O. Because of the burstiness and the asynchronicity of memory operations, the I/O may not be active all the time. Moreover, the main clock and the data lines may not be aligned due to skew. In such cases, additional strobe signals are used to indicate when to read and write the data.

[0040] The interface between a system memory controller and multiple Dynamic RAM devices has been well optimized over multiple design generations for both transfer speed and low power consumption. As one example, the present state of the art DRAM interface, LPDDR4, is comprised of 8 data lines, 1 DMI signal, 2 strobe lines, as well as other non-data-transfer related lines.

[0041] There is considerable interest in extending LPDDR4 to support higher performance at equal or less power consumption, but simple performance extrapolations of the existing technology seem problematic. Decreasing signal integrity precludes simply raising data transfer rates using the existing single-ended interconnection, and misalignment of received DRAM data and its strobe signal is a known issue even at current clock speeds. However, introduction of new technology is constrained by a strong desire to retain as much of the conventional practice as possible regarding bus layout, signal distribution, clocking, etc., as well as a hard requirement that the new technology be implementable in both the high-speed CMOS process used for memory controllers, and in the highly specialized DRAM fabrication process which produces extremely small, high capacitance and low leakage memory cells, but comparatively slow digital and interface logic.

[0042] Because of this slow logic speed, conventional DRAM designs utilize two or more phases of processing logic to handle the current LPDDR4 data transfer rates, as one example using one phase of processing logic to capture data on the rising edge of the data transfer strobe, and another phase of processing logic to capture data on the falling edge of the strobe. One hidden limitation of such multi-phased processing embodiments is the difficulty of extracting difference-based information from consecutively received unit intervals, as consecutive unit intervals by definition are known only by different processing phases. Thus, multi-phased processing is problematic for codes using transition-encoding, as well as for embedded- or self- clocking data solutions that rely on comparison of data values received in consecutive unit intervals.

[0043] These issues of clock extraction, and transition- or change-detection are most intractable in the communications receiver embodiment, thus the examples herein focus on embodiments in which a relatively slow DRAM device is the receiver and the more complex clock timing operations are performed by the transmitting controller device. No limitation is implied, as one familiar with the art will readily acknowledge that bidirectional data communication, including with DRAM devices, is well understood. It is equally obvious that the described embodiments may also be implemented using less constrained integrated circuit processes, including

conventional high speed integrated circuit processes and nonvolatile memory processes. DRAM embodiments may choose to utilize known art transmit clocking, essentially "turning around" the received clock and using it as an in-phase transmit clock with the controller's receiver performing the necessary clock phase adjustment. Alternatively, DRAM embodiments may incorporate local clock generation and/or clock phase adjustment as taught herein, or in combination with known art methods.

Receivers using Multi-input comparators

[0044] As described in [Holden I], a multi -input comparator with coefficients a 0 , ci \ , ... ,a m .\ is a circuit that accepts as its input a vector (x 0 , x\, ..,x m -\) and outputs

Result = [ao*xo + ...+ a m -i*Xm-i ) (Eqn. 1)

[0045] In many embodiments, the desired output is a binary value, thus the value Result may be sliced with an analog comparator to produce a binary decision output. Because this is a common use, the colloquial name of this circuit incorporates the term "comparator", although other embodiments may use a PAM-3 or PAM-4 sheer to obtain ternary or quaternary outputs, or indeed may retain the analog output of Eqn. 1 for further computation.

[0046] [Holden 1} and [Ulrich I] teach that the ENRZ may be detected using three instances of a four input multi-input comparator, performing the operations

Ro = (A + C) - (B + D) (Eqn. 2)

Ri = (C + D) - (A + B) (Eqn. 3)

R 2 = (C + B) - (D + A) (Eqn. 4)

or their algebraic equivalents, optionally including a normalization or scaling factor on the results. These operations may be readily performed with three identical instances of a multi-input comparator with coefficients of [ +l +1 -1 -1 ] and distinct permutations of the four input values as described in Eqn. 2 - 4.

[0047] In general, embodiments of such multi-input comparator receivers operate asynchronously up to the actual capture of results. The boundary between

asynchronous and clocked processing domains then varies, with some embodiments subsequently produce digital outputs using level-detecting comparators and thus deferring entry to a clocked processing domain, as illustrated in FIG. 2. Other embodiments utilize clocked samplers to constrain the measurement operation on the MIC outputs in both amplitude and time dimensions, creating at the sampling point a transition to clocked operation. To avoid confusion, the term "comparator" will be used herein to describe a measurement constrained only in amplitude, while

"sampler" will describe a measurement also constrained in time as with a sampling clock.

ODVS sub-channels

[0048] The block diagram of FIG. 2 provides a system-level overview for the subsequent examples illustrated as FIGs. 4-6. As with the previous MIC example, ENRZ is used herein for descriptive convenience, with no limitation implied. As the primary differences among these examples are in internal operational details of the transmitter with all examples producing equivalent transmit data streams, a receiver as illustrated in FIG. 2 may be assumed.

[0049] It is conventional to consider the data input to an ODVS encoder as vectors of data (i.e., a data word) to be atomically encoded as a codeword to be transmitted across a communications channel, detected by a receiver, and ultimately decoded to produce a received reconstruction of the transmitted vector or data word. An embodiment of a transmitter compatible with this model is shown in FIG. 4, wherein a data word So, Si, S2 IS accepted by codeword encoder 412, producing a codeword 414. As the symbols comprising the data-modulated ENRZ codeword are to be

communicated over discrete wires of communications channel 120, each such symbol is converted into suitable signaling levels 416 and emitted via line drivers 418 to the communications channel 120. ENRZ utilizes a four letter alphabet, generally represented by the values { +1, +1/3, -1/3, -1 } . Thus, one practical embodiment of an ENRZ transmitter as shown in FIG. 4 uses two binary signal wires per symbol or a total of eight wires to connect the ENRZ encoder to the signaling level

converter/output buffer driving four output wires; other embodiments may use different internal coding including without limitation analog signal levels, different numbers of binary signals, etc. [0050] The ENRZ code contains eight unique codewords, consisting of the permutations of { +1 , -1/3, -1/3, -1/3 } and {-1, 1/3, 1/3, 1/3} , which are sufficient to uniquely encode three binary bits. One embodiment of encoder 412 uses a simple look-up table to map input word So, Si, S2 to a data-modulated codeword. Equivalent embodiments utilize Boolean logic to perform the same operation at higher speed than possible using a memory-based look-up.

[0051] It is possible to design codeword encoders that operate asynchronously, using only unclocked Boolean logic devices. Other designs may rely on clocked pipelining or parallel phases of processing. What all codeword encoders have in common, however, is that their look-up or Boolean logic computation relies on multiple input values being simultaneously present in producing an output codeword. That is, an output codeword is a monolithic data object, dependent on more than one input value. Thus, it follows that if the output of a codeword encoder must resolve to a stable value quickly, its input values should only transition concurrently, thus implying they are within the same clock domain (as that term is conventionally understood in the art). It further follows that when operating at speeds limited primarily by the encoding operation's latency (a common situation in practice) a prudent system embodiment will present the encoder with inputs at one clock tick and accept the resultant codeword at a subsequent clock tick (i.e. synchronously).

[0052] It is equally accurate to model an ODVS communications system in a different way. As originally described in [Cronie I], an ODVS code may be fully specified by a Matrix having certain well-defined qualities; in particular, such a Matrix has an initial row consisting of all 1 values, with each subsequent row consisting of values summing to zero and being orthogonal to all other rows. Columns of the matrix correspond to signals on particular communication channel wires, while rows of the matrix correspond to mutually orthogonal subchannels capable of carrying information. In practical embodiments the first row, corresponding to communication over the common mode of the wires, is not used for transmission.

[0053] FIG. 5 is a block diagram of an embodiment based on this model. Modulation of a given subchannel corresponds to multiplication of its matrix row values by the modulating information signal, as performed by Subchannel encoders 512. Depending on the particular embodiment and the characteristics of the Matrix describing the subchannels, some or all of Subchannel encoders 512 may represent trivial functions such as assignment of an input value scaled by certain weighting values to certain outputs, or may represent operations performed by Boolean logic, or by digital lookup tables (LUT.) All modulated subchannel result vectors as in 513 are summed in an asynchronous codeword generator 514. By convention, the vector of elements comprising this summation result, also known as an asynchronous-transmit codeword, is described as being normalized to the range [-1, +1] for mathematical consistency, although in some physical embodiments this normalization is subsumed into other operations rather than being performed independently. The resulting vector of elements is converted into physical signaling levels 516 and output 518 as analog physical signals onto the communications channel wires 120.

[0054] Equivalent and compliant embodiments may utilize circuits or subsystem that combine one or more of the described steps. As one example, at least one embodiment integrates conversion 516 and output buffering 518 operations in a combined subsystem.

[0055] Such subchannel encoding is further illustrated in the flow chart of FIG. 11. A first input signal is received by a first subchannel encoder 1 110, which responsively generates elements of a first weighted subchannel vector 1 115. A second input signal is received by a second subchannel encoder 1 120, which responsively generates elements of a second weighted subchannel vector 1125. Asynchronous codeword generator 1 130 adds the first and second weighted subchannel vectors 1 115 and 1125 to produce asynchronous-transmit codeword output 1 135, wherein elements of 1135 transition asynchronously in response to transitions of 11 15 and/or 1 125. The elements of asynchronous-transmit codeword 1 135 are transmitted 1130 as multilevel analog signals 1 145 on a multi-wire bus or communications channel.

[0056] As previously described, in various embodiments the subchannel vectors may be generated by analog computation, by digital encoding as sets of one or more bits representing elements of the subchannel vectors, such digital encoding being performed by Boolean digital logic, look-up tables, or other embedded digital computational elements. The summation of the subchannel vectors to generate an asynchronous-transmit codeword may similarly be performed by analog computation, by digital computation using digital adders, Boolean logic circuits, look-up tables, or other embedded digital computational element.

[0057] In some embodiments, changes or transitions of value of the first input signal and the second input signal may occur asynchronously, synchronously with identical or nonidentical phase, or plesiochronously with varying or nonidentical phase. In at least one such embodiment, the first input signal is a data signal and the second input signal is a clock signal. In some embodiments, transitions on the second input signal occur one half transmit unit interval delay relative to transitions on the first input signal. In other embodiments, the first input signal transitions at a first rate, and the second input signal transitions at a second rate that is an integer fraction of the first rate.

Equivalence of the codebook and subchannel models

[0058] As described in [Cronie I], the H4 or ENRZ code is defined by Hadamard matrix of order 4. Thus, three rows of the matrix represent usable communication subchannels, and the four columns represent signals to be carried over the four wires of the communication channel. Enumerating all possible combinations of binary values modulating each of the subchannels gives 2 3 distinct summed subchannel results, equivalent to the eight "codewords" of the previous model.

[0059] One familiar with the art may observe that the vectors of input weights for multi-input comparators which a compatible receive embodiment may use to detect the ENRZ code is similarly described by rows of the matrix; as with the transmitter, practical embodiments do not implement a MIC corresponding to the first row of the matrix representing common mode transmission. As all other rows of the matrix are orthogonal to that first row, all MICs corresponding to those other rows will thus have inherent common mode immunity.

[0060] Communication over ODVS channels need not be limited to binary modulation. [Shokrollahi IV] teaches that PAM-3 and higher-ordered codes may be used over ODVS subchannels, and [Shokrollahi V] teaches that the modulation amplitude of a given ODVS subchannel may be adjusted, so as to provide comparable adjustment of the corresponding MIC output level at a receiver, thus demonstrating that a subchannel acts as a linear communications medium. However, as both [Shokrollahi IV] and [Shokrollahi V] presume conventional codeword-based communication models, they caution that such modulations can significantly impact the alphabet size required to describe the resulting codewords (and thus also the number of discrete signal levels appearing on communication channel wires,) and describe methods for constraining such alphabet growth.

[0061] Interpreting the system embodiment illustrated in FIG. 2 using this subchannel model, input data vector 100 entering communications transmitter 110 is shown as being comprised of individual bits So, Si, S2 entering encoder 112. Each of the individual bits So, Si, S2 modulate a separate subchannel, corresponding to multiplication of the input value by its corresponding orthogonal vector of the matrix defining that ODVS code. The resulting signals emitted onto the wires are thus the superposition (i.e. the summed values) of the resulting modulated subchannels.

[0062] The internal structure of receiver 132 is comprised of four receive front ends (as 131) that accept signals from wires 125, and optionally may include amplification and equalization as required by the characteristics of the communications channel 120. Three multi-input comparators are shown with their inputs connected to the four received wire signals as described by Eqns. 2, 3, and 4. For avoidance of confusion, the multi-input comparators are illustrated as being comprised of a computational function 233 performing the operations of Eqn. 2, Eqn. 3, and Eqn. 4 resulting in analog outputs representing the corresponding modulated subchannel signals, followed by a slicing function 234 producing digital outputs Ro, Ri, R2 corresponding to the binary modulation values So, Si, S2 accepted by the transmitter. No limitation to this configuration is implied, however, as in practice these functions may also be combined together or with other circuit elements to provide equivalent functionality.

Independence of subchannel timing

[0063] One familiar with the art may note that the ODVS encoder described in this manner is not constrained to produce a particular combination of wire outputs (e.g. in the previous encoding model, a single codeword) for each transmit unit interval. The familiar communications concept of "unit interval" may continue to have validity with respect to maximum signaling rate over any given subchannel, but not necessarily with respect to timing limitations between modulation on two different subchannels. In particular, this implies that the input data in this subchannel model need not be constrained to a single clock domain.

[0064] As an example, consider an instance of this embodiment not utilizing any timing-based component such as clocked latch or clocked multiplexor in its subchannel encoders nor in its MIC-based receivers. One may then observe, as one specific example offered without limitation, the input data bit S2 may change state one half unit interval later than the input data bits So and S i entering the transmit encoder and, as long as such state changes do not occur on a given subchannel more frequently than once per unit interval, the aggregate signaling capacity of the communications channel will not be exceeded and the receiver may detect all three results without error, reproducing at its output the same timing relationship as the input values had when presented to the encoder. Interpreting this result in the context of conventional codeword encoding, an asynchronous codeword encoder produces modulated codeword outputs based on the state of its inputs. Thus, input data bits So and S i and a first state of input bit S2 will result in a first modulated codeword output, and when input bit S2 asynchronously transitions to a second state, a new modulated codeword will be output derived from the new combination of inputs So, S i. and S 2 .

[0065] One of exceptional knowledge in the art may note that the timing relationship at the receiver may not be exactly duplicate that of the transmitter, as the three subchannels may represent distinct propagation modes of the communications medium having slightly different propagation velocities. Similarly, physical variations among elements of the transmitter and receiver may introduce timing skews between the subchannel signals impacting the timing relationship. However, in a given practical embodiment it is reasonable to assume that such variations are both small and consistent, thus capable of being addressed using well known practices of the art.

[0066] Continuing the example, the same embodiment may use, for example, the S2 subchannel to communicate a reference clock signal from transmitter to receiver, the receiver using transition edges of S2 to latch received data values So and Si at the optimum (i.e. center of "eye") time. This desired clock-to-data phase relationship is produced and controlled at the transmitter, with no elaborate PLL, DLL, or adjustable timing delay function being required at the receiver. The required timing relationship at the transmitter may be determined by design, calculation or estimation; in a further embodiment, the actual timing relationship and/or other receiver characteristic such as error rate may be measured at the receiver, communicated via a return channel to the transmitter, and used to adjust or correct the clock phase relative to data transitions.

[0067] FIG. 3 is a block diagram of a system embodiment incorporating such a transition-offset reference clock. The transmitter accepts two binary data inputs and a square wave clock signal making exactly one transition per unit interval (herein subsequently described as a half-rate clock.) In one embodiment used for purposes of illustration, the half-rate clock transitions simultaneously with transitions of the binary data inputs, and a ½ UI delay 310 is then introduced prior to encoder 312, which encodes the two data values and the phase-offset clock on a first, second, and third subchannels, respectively.

[0068] FIGs. 10A and 10B show example embodiments of delay 310 utilizing, respectively, a delay line utilizing logic delay elements and a phase interpolator.

[0069] In FIG. 10A, multiple delay buffer elements 1010, 1011, 1012, 1013, and 1014 introduce propagation delay into input signal Tx Clock, resulting in output Phase Delayed Tx Clock. The individual and/or aggregate delays of the delay buffer elements may in some embodiments be adjusted by modulation of an implementation parameter such as supply current, node capacitance, path resistance, or other element or behavior known in the art. Such adjustment may be performed one time, as at initialization, or during operation as part of a management or closed loop control behavior as known in the art. acting upon multiple phases from a clock generator.

[0070] The embodiment of FIG. 10B utilizes (in this example) two distinct phases of input signal input to Phase Interpolator circuit 1020, which outputs Phase Delayed Tx Clock having a intermediary phase. As is well known in the art, more than two distinct phases of input signal may be utilized, with the Phase Interpolator selecting among said inputs as well as interpolating between them.

[0071] In an equivalent embodiment, the clock generator producing the data clock and the half-rate clock may be designed to introduce a fixed 90 degree phase offset into the half-rate clock output, as will be well understood to a practitioner of the art. [0072] Receiver 130 detects the subchannels, producing received data 345 corresponding to received information on the first and second subchannels, and received clock 346 corresponding to received information on the third subchannel. Positive transitions of clock 346 trigger data latch 360 and, due to the action of inverter 350, negative transitions of clock 346 trigger data latch 370, producing latched data outputs 380 and 385. Because of the ½ UI phase offset introduced at the transmitter, the receiver may latch data with optimum "center of receive eye" timing, without need of receive-side delay components.

[0073] The example receiver subsystem 390 of FIG. 3 is further illustrated in the example embodiment of FIG. 9 incorporating multi-phase receive processing as may be utilized in DRAM designs. In such an embodiment, the clocked flip-flops (810- 813 as examples) available do not operate quickly enough to capture received data on consecutive unit intervals.

[0074] Operation of the receiver is as previously described, with signals received from interconnection 120 detected by receiver 130 which incorporates Multi Input Comparators 232, resulting in detected subchannels carrying signals Datai-DataN 345 and clock signal Rx Clock 346. To capture signals Datai-DataN at full speed, four distinct sets of flip-flops 810, 811 , 812, 813 are used as Data Latches clocked by non- overlapping and sequentially activating clocks CkO, Ckl, Ck2, and Ck3 respectively, causing data to be latched by latches 810 on the first rising edge of Rx Clock, by 81 1 on the first falling edge of Rx Clock, by 812 on the subsequent rising edge of Rx Clock, and by 813 on the subsequent falling edge of Rx Clock. The clock signals are steered in this manner by the combination of Flip-Flop 820 and AND gates 840, 841 , 842, 843. OR gate 830 causes Flip-Flop 820 to toggle after activation of Ckl and Ck3, insuring that the sequence will continue as described. Numerous other equivalent embodiments are known in the art, as are embodiments utilizing fewer or greater numbers of data latch phases, which may be applied in any combination with the described embodiments.

Hybrid embodiments

[0075] For descriptive purposes, this document will use the term "independent subchannel" for encoder embodiments of the type shown in FIG. 5, absent any timing constraints (e.g. such as clocked latches) that would impact the arbitrary phase relationship among communicated signals as described.

[0076] One must acknowledge that a number of useful implementation techniques become unavailable in a purely asynchronous embodiment as described above. Thus, it is useful to consider what will hereafter be called "hybrid subchannel"

embodiments, in which some portion of the data path is clocked, while other portions are asynchronous (i.e. representing multiple clock domains.)

[0077] In one embodiment utilizing such a configuration illustrated in FIG. 6, the codeword encoder 610 processing the data will output a data-modulated codeword 613 representing a weighted summation of the data modulated subchannels, absent the subchannel used for clock information. The subchannel used for clock information is separately encoded using an independent subchannel encoder 512 producing a clock- modulated codeword 513, the data-modulated codeword and the clock-modulated codeword being added 514 in an asynchronous codeword generator configured to generate an asynchronous-transmit codeword 515. Symbols of this asynchronous- transmit codeword are then converted to analog signal values 516 and output 518 as analog physical signals onto a multi-wire bus as previously described relative to FIG. 5. Other embodiments may combine elements, as combining summation 514, conversion 516, and output 518 in an output subsystem such as an output driver. In an alternate embodiment utilizing such a configuration, the digital encoder processing the data 610 is configured as if the data input corresponding to the subchannel used for the clock information were fixed at a constant value, e.g., zero. The clock modulated codeword 513 being added (as one example, within said output driver performing the combined operations 514, 516, 518) is then configured to represent differences from the fixed or default data-modulated subchannel values already present in that first result.

[0078] It will be readily apparent to one familiar with integrated circuit design that other embodiments providing such hybrid summed channel operation are possible: Data values representing subsets of all available subchannels and encoded as codewords using clocked logic may be converted to analog values which are then summed with analog values derived from unclocked encoding of other subchannels of data and/or clocks having arbitrary phase relationships. For purely digital embodiments, the alphabet representing values in a codeword and/or the values representing an asynchronously encoded clock or data subchannel may be encoded using a digital counting sequence immune to timing-related glitches, such as a grey code, one-hot code, or other digital encoding known in the art. Such sequences can be combined and used to generate a summed output value using well known digital-to- analog methods known in the art, with minimal risk of generating an incorrect output value because one or both of the input values was changing during the combining operation. In an alternative embodiment, known art "clock free" or asynchronous logic design methods may be used solely within the final combining element, producing a digital summation of the subchannel input values regardless of their source timing relationships. In a further alternative embodiment, the well known method of capturing the two input values in retiming latches to eliminate metastable or transient input states is applied, prior to conventional summation and conversion to an output value suitable for driving the output wires.

Hybrid subchannel combination in an output driver

[0079] The Glasswing embodiment described in [Hormati I] utilizes an ODVS code to communicate five data bits over a six wire communications channel, further utilizing a two wire communications channel to communicate a reference clock signal. Both the transmitter and receiver incorporates multiple parallel processing phases in its data path to achieve exceptionally high data rates, multiplexing into and out of these multiple phases being performed by high speed clocks. Thus, reimplementation of the Glasswing design as a purely asynchronous linearly summed subchannel embodiment would be quite difficult.

[0080] However, the line driver incorporated in at least one Glasswing embodiment is based on that of [Ulrich II], herein called an Ulrich-type driver, being comprised of multiple instances of relatively low power binary signal drivers configured to operate essentially in parallel, each such driver being capable of injecting a controlled amount of current into the common and resistively terminated output wire. The ODVS encoder is configured to output multiple binary control signals per codeword symbol, each such signal controlling different sets of line drivers within a given driver instance. Thus, a given codeword symbol may enable different sets and different numbers of paralleled line drivers, resulting in the production of different output signal levels for each codeword symbol.

[0081] The description of [Ulrich II] presumes that data-modulated codeword symbols are provided to the driver from multiple parallel processing instances combined by a clocked multiplexor. However, it is both possible and practical to reserve some number of Ulrich-type output driver instances and assign them to a separate signal source, one such embodiment incorporating an unclocked subchannel encoder instance modulated by a clock signal having an arbitrary phase relationship with the clocked data signals, as illustrated in FIG. 7. The illustrated embodiment accepts clock signal Hclk having an arbitrary phase relationship to data and/or codeword-encoded data transitions (as illustrated by the clock signal passing through phase delay 710), encoded as modulation of a single subchannel 720 and presented to N output driver slices or instances 750. Within each of said M instances, Driver Logic 730 controls each of the individual output drivers as 740 to produce a common output signal. The data path is unchanged from its description in [Ulrich II], with N output driver instances 760 being controlled by encoded data and Finite Impulse Response (FIR) variations of said encoded data.

[0082] Another such embodiment incorporating a clocked subchannel encoder instance operating in a different clock domain from the clocked data signals is illustrated in FIG. 8 (also derived from a drawing in [Ulrich II]) where N output driver slices are assigned to a clock-modulated subchannel encoded input, while M output driver slices remain assigned to a data-modulated codeword encoded input as in the original Ulrich design. For generality, each of the N slices is shown as including phase demultiplexing circuits driven by clocks sclk4 and sclk2, of the subchannel encoding clock domain. Conversely, each of the M slices includes phase

demultiplexing circuitry driven by clocks cclk4 and cclk2, of the codeword encoder clock domain.

[0083] As will be obvious to one familiar with circuit design, other embodiments may eliminate unnecessary demultiplexing because parallel clock processing phases are not necessary, unneeded FIR shaping is removed, etc. Regardless of such variations, the multiple paralleled instances of the output drivers driving common wire outputs effectively obtain the desired summation of the clock-modulated and data-modulated elements; e.g. the independent subchannel or separate clock domain modulated subchannel carrying the clock, and the codeword- modulated subchannels carrying the data.

[0084] It should be noted that any of the described hybrid solutions that generate an intermediate analog value prior to the actual output driver may require some form of linear amplification of that value to drive the output wires. Power consumption and integrated circuit area for such an amplifier may be significant in some embodiments. The described hybrid Ulrich-type driver avoids these issues by performing the combination within the output driver itself.

[0085] Similarly, hybrid solutions encoding the required output value digitally and then converting the result at the output, as one example using a D-to-A converter design capable of driving the considerable output load of the signal wires, may need to support a significant number of bits of resolution (e.g. a large alphabet of output values) so as to accurately reproduce the summation of all modulated subchannels regardless of the phase of their modulations.

[0086] The examples presented herein illustrate the use of vector signaling codes for point-to-point half-duplex or unidirectional wire communications. However, this should not been seen in any way as limiting the scope of the described embodiments, as the described methods and apparatus may be equally well applied to multi-drop and/or duplex communications environments. The methods disclosed in this application are equally applicable to other communication media including optical and wireless communications. Thus, descriptive terms such as "voltage" or "signal level" should be considered to include equivalents in other measurement systems, such as "optical intensity", "RF modulation", etc. As used herein, the term "physical signal" includes any suitable behavior and/or attribute of a physical phenomenon capable of conveying information. Physical signals may be tangible and non- transitory.