Abstract: A multichip package having a main die coupled to one or more daughter dies is provided. The main die may include embedded universal interface blocks (UIB) each of which can be used to interface with a corresponding daughter die to support high bandwidth parallel or serial communications. Each UIB may include an integrated processor subsystem and associated pattern sequencing logic to perform interface initialization and margining operations. Each UIB may perform simultaneous accesses to a daughter die across one or more channels. Each UIB may also include multiple phase-locked loop circuits for providing different clock signals to different portions of the UIB and a 2× clock phase generation logic. Each UIB may include multiple IO modules, each of which may optionally include its own duty cycle correction circuit. Each IO module may include buffer circuits, each of which may have a de-emphasis control logic for adjusting buffer drive strength.

Abstract: A multichip package having a main die coupled to one or more daughter dies is provided. The main die may include embedded universal interface blocks (UIB) each of which can be used to interface with a corresponding daughter die to support high bandwidth parallel or serial communications. Each UIB may include an integrated processor subsystem and associated pattern sequencing logic to perform interface initialization and margining operations. Each UIB may perform simultaneous accesses to a daughter die across one or more channels. Each UIB may also include multiple phase-locked loop circuits for providing different clock signals to different portions of the UIB and a 2× clock phase generation logic. Each UIB may include multiple IO modules, each of which may optionally include its own duty cycle correction circuit. Each IO module may include buffer circuits, each of which may have a de-emphasis control logic for adjusting buffer drive strength.

Abstract: A method of configuring a programmable integrated circuit device. A channel source within the virtual fabric is configured to receive input data from a first kernel outside of the virtual fabric and on the programmable integrated circuit device, and a channel sink within the virtual fabric is configured to transmit output data to the first kernel. The configuring of the channel source is modified such that the channel source receives input data from a second kernel in response to detecting a change in operation of the programmable integrated circuit device.

Abstract: Systems and methods are provided for supporting wide-protocol interface across a multi-die interconnect interface. Data signals of a wide-protocol interface are split into a plurality of data streams. A handshake signal is established between a first circuit and a second circuit, whereby the first circuit and second circuit are dies of a multi-die device. The first circuit transmits the plurality of data streams to the second circuit via a plurality of multi-die interconnect channels. Each data stream of the plurality of data streams are compressed based on the handshake signal in order to provide wide-protocol interface with reduced number of required pins.

Abstract: A multichip package having a main die coupled to one or more daughter dies is provided. The main die may include embedded universal interface blocks (UIB) each of which can be used to interface with a corresponding daughter die to support high bandwidth parallel or serial communications. Each UIB may include an integrated processor subsystem and associated pattern sequencing logic to perform interface initialization and margining operations. Each UIB may perform simultaneous accesses to a daughter die across one or more channels. Each UIB may also include multiple phase-locked loop circuits for providing different clock signals to different portions of the UIB and a 2× clock phase generation logic. Each UIB may include multiple IO modules, each of which may optionally include its own duty cycle correction circuit. Each IO module may include buffer circuits, each of which may have a de-emphasis control logic for adjusting buffer drive strength.

Abstract: Configurable specialized processing blocks, such as DSP blocks, are described that implement fixed and floating-point functionality in a single mixed architecture on a programmable device. The described architecture reduces the need to construct floating-point functions outside the configurable specialized processing block, thereby minimizing hardware cost and area. The disclosed architecture also introduces pipelining into the DSP block in order to ensure the floating-point multiplication and addition functions remain in synchronicity, thereby increasing the maximum frequency at which the DSP block can operate. Moreover, the disclosed architecture includes logic circuitry to support floating-point exception handling.

Abstract: A programmable integrated circuit may be provided with a memory interface for communicating with an external memory over a serial communications path. To accommodate a variety of different memory interface protocols while satisfying low-latency performance criteria, part of the memory interface may be formed from programmable logic and part of the memory interface may be formed from hardwired circuitry. The programmable logic of the memory interface may be used to implement packet formation logic that creates packets that include empty fields for sequence number information, acknowledgement information, and cyclic redundancy check information. The hardwired circuitry of the memory interface may be used to insert a sequence number, an acknowledgement, and cyclic redundancy check information into the empty fields.

Abstract: A method for implementing physical optimizations includes performing physical optimizations on a first reference version of a design, maintaining a computer-readable list of the physical optimizations, and during a subsequent compile for a second version of the design: identifying matching cells, nets, or both between the first reference version of the design and the second version of the design; and restoring at least a subset of the physical optimizations in the second version of the design by reading the computer-readable list of the physical optimizations and applying the subset to a computer-readable description of the second version of the design.

Abstract: A method for dynamically configuring multiple processors based on needs of applications includes receiving, from an application, an acceleration request message including a task to be accelerated. The method further includes determining a type of the task and searching a database of available accelerators to dynamically select a first accelerator based on the type of the task. The method further includes sending the acceleration request message to a first acceleration interface located at a configurable processing circuit. The first acceleration interface sends the acceleration request message to a first accelerator, and the first accelerator accelerates the task upon receipt of the acceleration request message.

Abstract: A method for implementing a convolutional neural network (CNN) accelerator on a target includes utilizing one or more processing elements to perform convolution. A configuration of the CNN accelerator is modified to change filters implemented by the CNN accelerator and to change formatting of output data. The one or more processing elements are utilized to perform one of deconvolution and backpropagation convolution in response to the change in the filters and formatting of the output data.

Abstract: Systems and methods for dynamically sizing inter-kernel communication channels implemented on an integrated circuit (IC) are provided. Implementation characteristics of the channels, predication, and kernel scheduling imbalances may factor into properly sizing the channels for self-synchronization, resulting in optimized steady-state throughput.

Abstract: Integrated circuits with specialized processing blocks are provided. The specialized processing blocks may include floating-point multiplier circuits that can be configured to support variable precision. A multiplier circuit may include a first carry-propagate adder (CPA), a second carry-propagate adder (CPA), and an associated rounding circuit. The first CPA may be wide enough to handle the required precision of the mantissa. In a bridged mode, the first CPA may borrow an additional bit from the second CPA while the rounding circuit will monitor the appropriate bits to select the proper multiplier output. A parallel prefix tree operable in a non-bridged mode or the bridged mode may be used to compute multiple multiplier outputs. The multiplier circuit may also include exponent and exception handling circuitry using various masks corresponding to the desired precision width.

Abstract: A multiple memory access system is disclosed. The system includes a first die disposed on a package substrate. A second die is stacked above the first die. The first die, the second die and the package substrate form a first package. An IC is placed within a close proximity of the first package where the first die communicates with the second die at a first data rate while the first die communicates with the IC at a second data rate. The first data rate is higher than the second data rate. Methods of forming a multi-access memory system are also disclosed.

Abstract: An integrated circuits with sequential logic circuitry is provided. The sequential logic circuitry may including latching circuits that receive clock signals from on-chip or off-chip clock sources. The clock signals may exhibit clock skew that is native to the integrated circuit. The natively existing clock skew can be leverage to perform time borrowing to help optimize circuit performance. The desired clock skew can be achieved by intelligent placement of the clock sources and deliberate routing of the clock signals from the clock sources to respective types of clock distribution networks on the integrated circuit.

Abstract: Circuits, methods, and apparatus for storing application data, keys, authorization codes, or other information in a volatile memory on an FPGA. A field programmable gate array (FPGA) can include multiple memory blocks and partition those blocks among multiple independent reconfigurable regions. Access to the memory blocks can then be restricted so that only authorized regions have access to particular memory partitions. In addition, each partition can store multiple message authentication codes (MACs) for further controlling access to data in each partition.

Abstract: A method for implementing a convolutional neural network (CNN) accelerator on a target includes utilizing one or more processing elements to implement a standard convolution layer. A configuration of the CNN accelerator is modified to change a data flow between components on the CNN accelerator. The one or more processing elements is utilized to implement a fully connected layer in response to the change in the data flow.

Abstract: A virtualization platform for Network Functions Virtualization (NFV) is provided. The virtualization platform may include a host processor coupled to an acceleration coprocessor. The acceleration coprocessor may be a reconfigurable integrated circuit to help provide improved flexibility and agility for the NFV. The coprocessor may include multiple virtual function hardware acceleration modules each of which is configured to perform a respective accelerator function. A virtual machine running on the host processor may wish to perform multiple accelerator functions in succession at the coprocessor on a given data. In one suitable arrangement, intermediate data output by each of the accelerator functions may be fed back to the host processor. In another suitable arrangement, the successive function calls may be chained together so that only the final resulting data is fed back to the host processor.

Abstract: Techniques to operate circuitry in an integrated circuit are provided. The circuitry may include rate detection circuitry, receiver circuitry, and configuration circuitry. The receiver circuitry may receive a data stream with an arbitrary data rate. The rate detection circuitry may receive a reference clock signal that is associated with the received data stream. The rate detection circuitry determines the frequency of the reference clock signal such that an appropriate clock signal may be generated for the receiver circuitry. The receiver clock signal may be generated by clock generation circuitry that is coupled to the rate detection circuitry. The configuration circuitry may accordingly configure the receiver circuitry based at least on the determined frequency of the reference clock signal so that the receiver circuitry may operate at the arbitrary data rate.

Abstract: In one embodiment, a computer-implemented method includes performing retiming using a first circuit design to determine one or more variations to the first circuit design, identifying one or more source registers that are involved in one or more unsuccessful retiming moves due to initial conditions conflicts, and recommending, via a graphical user interface, one or more corrective actions to the one or more source registers to avoid the initial conditions conflicts and improve performance of the one or more variations.

Abstract: Configuration data for an integrated circuit may be generated using logic design equipment to implement a logic design on the integrated circuit. The equipment may perform multiple rounds of incremental physical synthesis, incremental timing analysis, and incremental legalization operations. Each round may involve performing multiple different physical synthesis transforms on the design that are individually rejected until transforms that satisfy legality constraints and improve timing for the logic design are found and incorporated into the netlist. The configuration data may then be generated using the netlist. In this way, the logic design may be incrementally altered and verified during the physical synthesis process. This prevents the need for rejecting or accepting an entire batch logic changes to the netlist even when only some of the changes are non-ideal, thus optimizing circuit performance as well as the compile time required to implement the logic design on the integrated circuit.