Chiplet gearbox for low-cost multi-chip module applications

[0001] This application is a Non-Provisional that claims priority to U.S. Provisional Application No. 63/295,524, filed Dec. 31, 2021, entitled INNOVATIVE LOW-COST PACKAGING SOLUTIONS FOR SYSTEM IN PACKAGE AND THE HBM CROSSOVER PHY ARCHITECTURE, U.S. Provisional Application No. 63/296,162, filed Jan. 4, 2022, entitled AN EFFICIENT GEARBOX ARCHITECTURE FOR CONVERSION OF HBM PHY INTERFACE TO A SIMULTANEOUS BIDIRECTIONAL PHY, and U.S. Provisional Application No. 63/341,897, filed May 13, 2022, entitled SIMULTANEOUS MEMORY READ AND WRITE FOR HIGHER BEACHFRONT BANDWIDTH EFFICIENCY, all of which are incorporated herein by reference in their entirety.

[0002] The disclosure herein relates to semiconductor devices, packaging and associated methods.

[0003] As integrated circuit (IC) chips such as system on chips (SoCs) become larger, the yields realized in manufacturing the chips become smaller. Decreasing yields for larger chips increases overall costs for chip manufacturers. To address the yield problem, chiplet architectures have been proposed that favor a modular approach to SoCs. The solution employs smaller sub-processing chips, each containing a well-defined subset of functionality. Chiplets thus allow for dividing a complex design, such as a high-end processor or networking chip, into several small die instead of one large monolithic die.

[0004] While beneficial in improving the yield issue for larger die, chiplet-based architectures for multi-chip modules (MCMs) often employ costly packaging components and associated processes.

[0014] Semiconductor devices, packaging architectures and associated methods are disclosed. In one embodiment, a multi-chip module (MCM) is disclosed. The MCM includes a first integrated circuit (IC) chip including a first interface defining a first number of interface contacts. Conversion circuitry receives a first full set of signals associated with the first interface and to omit a subset of the full set of signals to generate a reduced set of signals. Serialization circuitry serializes the reduced set of signals to generate a serialized set of signals. A second interface transmits the serialized set of signals with a second number of interface contacts that is less than the first number of interface contacts. A logic IC chip includes a third interface coupled to the second interface via a set of links and configured to match the second interface. Deserialization circuitry deserializes the serialized set of signals. Reconversion circuitry recreates signals corresponding to the omitted subset of the full set of signals and aggregates the recreated signals with the deserialized signals to form a second full set of signals that correspond to the first full set of signals. By omitting a subset of the full set of signals to generate a reduced set of signals, then later recreating signals corresponding to the omitted subset of the full set of signals, an efficient reduction in the number of interface bumps can be achieved, allowing for the use of larger and less expensive bumps for use with a fewer number of signaling links. Other embodiments utilize efficient error coding techniques, such as forward error correction (FEC) to address bit errors associated with the larger but less-costly interface bumps.

[0015] Throughout the disclosure provided herein, the term multi-chip module (MCM) is used to represent a semiconductor device that incorporates multiple semiconductor die or sub-packages in a single unitary package. An MCM may also be referred to as a system in a package (SiP). With reference to FIG. 1, a block diagram of one embodiment of a multi-chip module (MCM) is shown, generally designated 100. For one embodiment, the MCM includes a package substrate (not shown) that forms a support surface for multiple integrated circuit (IC) chips, or chiplets, including a first IC chip 102, a second IC chip 108 and a third IC chip 120. For some applications, the package substrate takes the form of a low-cost non-silicon substrate.

[0016] With continued reference to FIG. 1, for one embodiment the first IC chip 102 takes the form of an IC memory device, such as a volatile dynamic random access memory (DRAM) device or a non-volatile Flash memory device. For one specific embodiment, the first chiplet takes the form of a DRAM memory device compliant with the High Bandwidth Memory (HBM) architecture. For HBM-related embodiments, the first IC chip may take the form of a logic interface chip or a bottom-most DRAM memory chip of a stack of memory chips. The first IC chip 102 includes an electrical interface, such as a memory interface 104 having various input/output (I/O) circuits (not shown) for communicating with other IC devices. The memory interface 104 couples to a mechanical interface in the form of an array of contacts 106, often referred to as bumps that electrically connect to one or more other corresponding bumps of a different device.

[0017] For one specific embodiment, and further referring to FIG. 1, the array of bumps 106 employed by the first IC chip are arranged in a manner that is compliant with the given signaling standard associated with the device. As an example, the HBM standard supports sixteen memory channels, with one-hundred twenty pins or I/O paths dedicated to support each channel. FIG. 2 illustrates a table that shows the various functions, number of bumps, and signal identifiers for each of the signals associated with a single channel of an HBM pinout. With sixteen channels, nineteen-hundred twenty bumps are employed in the bump array for an HBM memory device. With surface area and overall footprint constraints, forming almost two-thousand bumps on the surface of an IC memory device for coupling to the required I/O paths is generally accomplished using high-density micro-bump packaging technology. While this packaging technology successfully aggregates micro-bumps at a very high density, associated yield rates and signal integrity limitations may involve additional packaging techniques that may increase cost to the overall multi-chip module 100.

[0018] Further referring to FIG. 1, to reduce costs associated in manufacturing the MCM 100, for one embodiment the first IC chip 102 is coupled to a second IC chip 108, referred to herein as a gearbox chiplet. For one embodiment, the gearbox chiplet 108 includes a second bump array 110 that matches the first bump array 106 of the first IC chip 102. For some embodiments, the two bump arrays 106 and 110 may be directly bonded to each other such that the first IC chip 102 and the second IC chip 108 form a stacked or 3.0D configuration for mounting on the package substrate.

[0019] In some embodiments, the features of the gearbox chiplet 108 may be employed in a logic interface chip or DRAM memory chip of an HBM memory device. In other embodiments, the first IC chip 102 and the second IC chip 108 may be disposed horizontally to each other in a 2.5D configuration, and coupled to each other via a secondary substrate (not shown) that is mounted on the package substrate. Various embodiments and implementations of the connections between the first chiplet and the second chiplet are shown and described in copending U.S. application Ser. No. 17/973,905, filed Oct. 26, 2022, entitled “Method and Apparatus to Reduce Complexity and Cost for Multi-Chip Modules (MCMs)”, assigned to the assignee of the present disclosure and incorporated by reference herein in its entirety.

[0020] With continued reference to FIG. 1, the gearbox chiplet 108 includes a first interface converter 112 that generally omits selected signals transferred from the first IC chip 102 in a receive direction to generate a reduced set of signals N, and recreates previously omitted signals in a transmit direction to the first IC chip 102. Serialization circuitry 114 serializes the reduced set of signals in the receive direction by a selected integer divisor M.

[0021] Further referring to FIG. 1, the gearbox chiplet 108 includes a third interface bump array 116 that takes advantage of the reduced number of signal paths associated with the serialized set of signals and employs a standard array of bumps that each exhibit a significantly larger size, pitch and overall dimension than the micro-bump array. The use of the low-density standard array bumps improves the reliability of the mechanical interface connections and avoids many of the signal integrity issues associated with high-density microbumps for high-speed applications. Although the standard array bumps are each larger than their micro-bump counterparts, the reduction in paths due to the first pass filtering and the second-pass serialization results in a significant path reduction that is supported by a similarly reduced number of bumps. As a result, the reduced number of larger bumps may be contained within a similar surface area as that dedicated to a much larger set of microbumps.

[0022] Further referring to FIG. 1, the reduction in signaling paths, and the use of the sparser standard array of bumps allows for the use of high-speed links, such as at 118, routed along the non-silicon substrate and for coupling the gearbox chiplet 108 to the third chiplet 120. For some embodiments, the links are high-speed short or long reach links that provide bandwidth capabilities that are similar to links incorporated by silicon substrates. For some embodiments, the links may take the form of unidirectional or bidirectional signaling links that may simultaneously communicate in both upstream and downstream directions. Further detail associated with one embodiment of a simultaneous bidirectional link is provided below.

[0023] With continued reference to FIG. 1, the third chiplet 120 includes a fourth interface bump array 122 that takes advantage of the serialized set of signals transferred along the high-speed links 118, and employs a second standard array of bumps that substantially matches the third interface array of bumps 116 formed on the gearbox chiplet 108. For the receive direction, the third IC chip 120 employs the deserialization circuitry 124 to convert the serialized signals from the high-speed links 118 back to a reduced parallel set of signals corresponding to the pre-serialized signals received by the gearbox chiplet 108. The second interface converter 126 then recreates and inserts bits corresponding to the omitted bits that were filtered by the gearbox chiplet 108 to re-form the full set of signals transmitted by the first IC chip 102. The full set of signals may then be fed to core circuitry 128 of the third IC chip 120. For HBM memory embodiments, the third IC chip 120 may take the form of a host chip or application specific integrated circuit (ASIC) processing chip that acts as a master chip with respect to the HBM memory chip (the first IC chip 102).

[0024] FIG. 3 illustrates further detail associated with one embodiment of the gearbox chip 108 of FIG. 1. The first interface converter 112 includes a first signal interface 302 that couples to the micro-bump array 106 of the first IC chip 102. For one specific HBM embodiment, the first signal interface reflects a full HBM memory interface, including sixty-four transmit data bits TxDQ and receive data bits RxDQ, and various command/address signals such as datamask DM, data bus inversion (DBI), read/write strobes RDQS/WDQS, row and column address bits, and others. In a receive direction, the first interface converter 112 employs bit removal circuitry 304 that selectively omits certain signals (when received from the HBM interface) to carry out a first-pass signal path reduction. In a transmit direction (for signals transmitted to the HBM interface), the first interface converter 112 includes bit insertion circuitry 306 to selectively recreate signals that were omitted in the receive direction. For one specific embodiment, and referring to FIG. 2, signal paths dedicated to the data bus inversion (DBI) function may be omitted via the bit removal circuitry 304, as well as redundant data paths, redundant address paths, and reserved for future use (RFU) paths. Additionally, several strobe paths (RDQS and WDQS) may be omitted. The effect of the first-pass signal path reduction may reduce the HBM path count, in a receive direction, from one-hundred twenty paths per channel to, for example, one-hundred paths per channel. With sixteen channels, the reduction in overall signaling paths reduces from nineteen-hundred twenty paths to sixteen-hundred paths. This per-channel path reduction is reflected in a second signaling interface 308 of the first interface converter 112.

[0025] Further referring to FIG. 3, the second signaling interface 308 of the first interface converter 112 is coupled to the first serialization/deserialization (SerDes) circuitry 114. The SerDes circuitry 114 includes multiple SerDes transceivers 312, each including a multiplexing transmitter 314 coupled to an input/output (I/O) pin 316, and a demultiplexing receiver 318 coupled to the pin 316. Each pin 316 couples to a given one of the high-speed links 118. For one embodiment, the multiplexing transmitters 314 may be configured to time-multiplex various incoming parallel signals into a serial stream of signals via a predetermined conversion ratio (such as 2:1, 3:1, 4:1, etc.) for transmission along the given link 118. This time-multiplexing feature provides a second-pass signal path reduction to further reduce the number of signal paths proportional to the conversion ratio. In an opposite signaling direction, the demultiplexing receivers 318 may be configured to demultiplex a serialized stream of signals into multiple parallel signals.

[0026] With continued reference to FIG. 3, for one embodiment, the high-speed links 118 are configured as simultaneous bidirectional links, that are each capable of simultaneously transmitting and receiving data via the SerDes circuitry 114. The use of the simultaneous bidirectional links provides a further technique to reduce the number of signal paths in the receive direction. A hybrid circuit (not shown) manages transmission and reception of data for transfer via each of the simultaneous bidirectional links 118. Other embodiments and associated details for various interface adapter chiplet configurations are described in U.S. patent application Ser. No. 16/812,234, titled “LOW COST SOLUTION FOR 2.5D AND 3D PACKAGING USING USR CHIPLETS”, filed Mar. 6, 2020, assigned to the Assignee of the instant application, and incorporated by reference in its entirety.

[0027] FIG. 4 illustrates further detail associated with one embodiment of the third IC chip 120 of FIG. 1. Generally, the third IC chip 120 includes interface circuitry that is the inverse of the gearbox chiplet 108. The third IC chip 120 employs the second SerDes interface 124 that couples to the high-speed links 118. The second SerDes interface 124 is formed similar to the first SerDes interface 114 of the gearbox chiplet 108, with demultiplexing receivers 404 to convert serial signal streams to parallel signals, and multiplexing transmitters 406 to convert parallel signals to serialized signals. The second interface converter 126 couples to the second SerDes interface 124 and includes second bit insertion circuitry 408 to recreate the signals that were omitted by the first interface converter 112 of the gearbox chiplet 108. Second bit removal circuitry 410 omits selected signals for destination to the first IC chip 102.

[0028] For one embodiment, the third IC chip 120 employs a forward error correction coder/decoder (FEC CODEC) 410 to provide error correction capability for bits stored in the memory, and those in transit via communications, thus improving the bit error rate (BER) of communications between the first IC chip 102, the gearbox chiplet 108 and the third IC chip 120. In certain embodiments, the FEC CODEC provides improved error correction capability without adding to the overhead of the memory channels. For one specific embodiment, instead of utilizing only a parity bit for every thirty-two bits of data, an FEC code is employed. In one embodiment, a Reed-Solomon code may be used to protect signals running at full speed, such as data DQ, error code signals ECC and SEV, which for an HBM embodiment total seventy-two signals. Since the ECC bits can be part of the FEC overhead, only sixty-eight signals need to be error protected. In an effort to generate a code having a relatively short frame size (to reduce latency), and exhibit a minimum overhead (such as coding circuitry overhead and/or bump size), one embodiment uses an RS code RS(72,68,8), which corrects two erroneous 8-bit words out of seventy-four words. To support this code, four ECC bits are utilized. In some embodiments, where a thirty-two bit sub-channel is used, the remaining thirty-two bit positions may be filled with known bits, such as all “0s”, or a given pattern such as 00001111. In such circumstances, sixty-four bits are still transferred, but with a relatively low bit toggling rate that consumes less power and results in a significantly lower error rate and which assists in the recovery of the actual thirty-two bits being transferred.

[0029] For one embodiment, and referring now to FIG. 5, the bump array pattern employed by the gearbox chiplet 108 to communicate with the third IC chip 120 employs a layout architecture that optimizes the number of bump contacts associated with a given gearbox conversion ratio for the available chiplet surface area. For one specific embodiment where the gearbox chiplet 108 utilizes a 3:1 conversion ratio, a bump pattern of DDPDDPDDPDDP is employed, with “P” bumps representing power/ground contacts, and “D” representing data contacts. For one specific embodiment, constraints used in supporting the bump pattern include limiting the dimensions of the gearbox chiplet 108 first signaling interface 110 (FIG. 1) to be no larger than the HBM interface 106, and targeting a bump pitch of 130 micrometers, an X dimension bump spacing of 65 micrometers, and a Y dimension bump spacing of 112.6 micrometers. For different conversion ratios, the bump pitches and X and Y bump spacings may be larger or smaller, depending on the conversion ratio and application.

[0030] FIG. 6 illustrates a further embodiment of a Table pinout for an HBM memory device, similar to Table 1 of FIG. 2, but showing respective input and/or output directions for each of the signals. For example, data signals including read data and write data, are generally bidirectional in nature and may be received by the memory device (in the case of write data), or transmitted by the memory device (in the case of read data). Various command and address signals, however, are generally unidirectional in nature. For some embodiments, simultaneous bidirectional links may be used to couple a memory device to a memory controller, with certain of the unidirectional command/address signal paths reused for read data signals to either further reduce the number of bumps or to increase the read bandwidth of the system.

[0031] FIG. 7A illustrates one embodiment of a memory system 700, including a memory device 702 coupled to a host memory controller 704 via a set of simultaneous bidirectional links 706. In a write direction, a first sub-set 708 of the links 706 may be employed for write data transfers, while a second subset 710 of the links 706 may be used for transferring command/address (C/A) signals from the controller 704 to the memory device 702. In a read direction, all of the links may be used to transfer read data simultaneous with the transfer of the write data and C/A signals. Consequently, in this embodiment, the read bandwidth is higher than the write bandwidth due to the availability of the extra sub-set of links for read data transmissions.

[0032] FIG. 7B illustrates an alternative embodiment of a memory system 720 that uses simultaneous bidirectional links, but with a reduced set of links in comparison to the system 700 of FIG. 7A. Rather than using sub-sets of links for the write direction, the embodiment of FIG. 7B time-multiplexes the command/address signals with write data along a full width of the links, and uses the same full width for read data transfers simultaneously with the transfers of the write data and C/A signals.

[0033] FIG. 8 illustrates one embodiment of a simultaneous bidirectional link 800 that may be used in the memory systems and MCM architectures described above. A first simultaneous bidirectional input/output (I/O) circuit 802 includes transmit circuitry 804 that transmits data to a link 806 and complement data to a summer 808. Receiver circuitry 810 receives data sent from a link partner I/O circuit 812 that has been summed with the complement data sent by the transmit circuitry 804, essentially cancelling any reflected transmit data signals that may be included in the receive data. The link partner I/O circuit 812 also includes transmit and receive circuitry similar to that of the local I/O circuitry.

[0034] When received within a computer system via one or more computer-readable media, such data and/or instruction-based expressions of the above described circuits may be processed by a processing entity (e.g., one or more processors) within the computer system in conjunction with execution of one or more other computer programs including, without limitation, net-list generation programs, place and route programs and the like, to generate a representation or image of a physical manifestation of such circuits. Such representation or image may thereafter be used in device fabrication, for example, by enabling generation of one or more masks that are used to form various components of the circuits in a device fabrication process.

[0035] In the foregoing description and in the accompanying drawings, specific terminology and drawing symbols have been set forth to provide a thorough understanding of the present invention. In some instances, the terminology and symbols may imply specific details that are not required to practice the invention. For example, any of the specific numbers of bits, signal path widths, signaling or operating frequencies, component circuits or devices and the like may be different from those described above in alternative embodiments. Also, the interconnection between circuit elements or circuit blocks shown or described as multi-conductor signal links may alternatively be single-conductor signal links, and single conductor signal links may alternatively be multi-conductor signal links. Signals and signaling paths shown or described as being single-ended may also be differential, and vice-versa. Similarly, signals described or depicted as having active-high or active-low logic levels may have opposite logic levels in alternative embodiments. Component circuitry within integrated circuit devices may be implemented using metal oxide semiconductor (MOS) technology, bipolar technology or any other technology in which logical and analog circuits may be implemented. With respect to terminology, a signal is said to be “asserted” when the signal is driven to a low or high logic state (or charged to a high logic state or discharged to a low logic state) to indicate a particular condition. Conversely, a signal is said to be “deasserted” to indicate that the signal is driven (or charged or discharged) to a state other than the asserted state (including a high or low logic state, or the floating state that may occur when the signal driving circuit is transitioned to a high impedance condition, such as an open drain or open collector condition). A signal driving circuit is said to “output” a signal to a signal receiving circuit when the signal driving circuit asserts (or deasserts, if explicitly stated or indicated by context) the signal on a signal line coupled between the signal driving and signal receiving circuits. A signal line is said to be “activated” when a signal is asserted on the signal line, and “deactivated” when the signal is deasserted. Additionally, the prefix symbol “/” attached to signal names indicates that the signal is an active low signal (i.e., the asserted state is a logic low state). A line over a signal name (e.g., ‘<signal name >’) is also used to indicate an active low signal. The term “coupled” is used herein to express a direct connection as well as a connection through one or more intervening circuits or structures. Integrated circuit device “programming” may include, for example and without limitation, loading a control value into a register or other storage circuit within the device in response to a host instruction and thus controlling an operational aspect of the device, establishing a device configuration or controlling an operational aspect of the device through a one-time programming operation (e.g., blowing fuses within a configuration circuit during device production), and/or connecting one or more selected pins or other contact structures of the device to reference voltage lines (also referred to as strapping) to establish a particular device configuration or operation aspect of the device. The term “exemplary” is used to express an example, not a preference or requirement.

[0036] While the invention has been described with reference to specific embodiments thereof, it will be evident that various modifications and changes may be made thereto without departing from the broader spirit and scope of the invention. For example, features or aspects of any of the embodiments may be applied, at least where practicable, in combination with any other of the embodiments or in place of counterpart features or aspects thereof. Accordingly, the specification and drawings are to be regarded in an illustrative rather than a restrictive sense.