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 method for designing a system on a target device includes performing speculative register retiming with speculative changes made to a design of the system after an initial compilation of the design. A strategy is generated for an actual register retiming in response to user specified preferences on the speculative changes.
Abstract:
A method for designing a system on a target device includes generating a solution for the system. A solution for a module of the system identified by a user is preserved. The preserved solution for the module is implemented at a location on the target device identified by the user.
Abstract:
In one embodiment, a tangible, non-transitory, computer-readable medium, includes instructions to receive a first circuit design, determine one or more variations of the first circuit design using register retiming with speculative circuit design changes, determine one or more performance improvements of the variations when fed clock signals over the first circuit design, determine one or more tradeoffs of the one or more variations of the first circuit design in comparison to the first circuit design, display a summary of the one or more variations of the first circuit design, the one or more performance improvements, and the one or more tradeoffs, and provide a user-selectable user interface element to enable selection of the first circuit design, at least one of the one or more variations of the first circuit design, or a combination thereof.
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 method for designing a system on a target device includes entering the system. The system is synthesized. The system is mapped. The system is placed on the target device. The system is routed. Physical synthesis is performed on the system immediately after more than one of the entering, synthesizing, mapping, placing and routing procedures.
Abstract:
A circuit design description may have registers and combinational gates. Circuit design computing equipment may perform register retiming on the circuit design description, whereby registers are moved across combinational gates, information about the register moves are recorded, and ultimately a modified circuit design description is created. The circuit design computing equipment may perform sequential equivalence checking to ensure that the circuit design description and the modified circuit design description are sequentially equivalent. To facilitate the sequential equivalence checking, the circuit design computing equipment may augment the two circuit design descriptions based on the information recorded during register retiming.
Abstract:
A circuit design may have registers and combinational gates. Circuit design computing equipment may perform register retiming in the circuit design, whereby registers are moved across combinational gates. Information about the register moves may be recorded, and a modified circuit design is created. The circuit design computing equipment may implement the circuit design in an integrated circuit. A logic analyzer may be used to debug the circuit design implemented in the integrated circuit in real-time and at high-speed. To facilitate the debugging process, the circuit design computing equipment may augment the integrated circuit and/or compensate for register retiming based on the information recorded during register retiming.
Abstract:
An integrated circuit may include memory interface circuitry for communicating with off-chip memory. The memory interface circuitry may receive data signals and data strobe signals from different memory devices via respective data ports and data strobe ports. The memory interface circuitry may be operable in at least first and second modes. In the first mode, data signals from each memory device may be received at two respective data ports while the data strobe signal from one memory device is used to clock the data signals at two corresponding read capture registers. In the second mode, data signals from first and second memory devices may be received via first and second data ports, respectively. The data strobe signal from the first memory device may be ignored while the data strobe signal from the second memory device is used to clock the data signals at two corresponding read capture registers.
Abstract:
A method for designing a system on a target device includes entering the system. The system is synthesized. The system is mapped. The system is placed on the target device. The system is routed. Physical synthesis is performed on the system immediately after more than one of the entering, synthesizing, mapping, placing and routing procedures.