Abstract: Implementing an application can include generating, from the application, a compact data flow graph (DFG) including load nodes, inserting, in the compact DFG, a plurality of virtual buffer nodes (VBNs) for each of a plurality of buffers of a data processing engine (DPE) array to be allocated to nets of the application, and, forming groups of one or more load nodes of the compact DFG based on shared buffer requirements of the loads on a per net basis. Virtual driver nodes (VDNs) that map to drivers of nets can be added to the compact DFG, where each group of the compact DFG is driven by a dedicated VDN. Connections between VDNs and load nodes through selected ones of the VBNs are created according to a plurality of constraints. The plurality of buffers are allocated to the nets based on the compact DFG as connected.

Abstract: An example integrated circuit (IC) die in a multi-die IC package, the multi-die IC package having a test access port (TAP) comprising a test data input (TDI), test data output (TDO), test clock (TCK), and test mode select (TMS), is described. The IC die includes a Joint Test Action Group (JTAG) controller having a JTAG interface that includes a TDI, a TDO, a TCK, and a TMS, a first output coupled to first routing in the multi-die IC package, a first input coupled to the first routing or to second routing in the multi-die IC package, a master return path coupled to the first input, and a wrapper circuit configured to couple the TDI of the TAP to the TDI of the JTAG controller, and selectively couple, in response to a first control signal, the TDO of the TAP to either the master return path or the TDO.

Abstract: An example a method of optimizing a neural network having a plurality of layers includes: obtaining an architecture constraint for circuitry of an inference platform that implements the neural network; training the neural network on a training platform to generate network parameters and feature maps for the plurality of layers; and constraining the network parameters, the feature maps, or both based on the architecture constraint.

Abstract: Circuits and methods for computing an order N polynomial include V decimation stages, each stage including respective multiply-and-accumulate circuitry. The multiply-and-accumulate circuitry in each stage k, in response to an input r-term and a plurality of input z-terms 0 through (Nk?1), generates output z-terms 0 through (Nk/2?1) and an output r-term as a square of the input r-term. Each output z-term i is a sum of input z-term (2i+1) of the input z-terms and a product of input z-term 2i and the input r-term. The multiply-and-accumulate circuitry in decimation stages k for k?(V?1) provides the output r-term and one or more output z-terms from decimation stage k as the input r-term and one or more input z-terms to the respective multiply-and-accumulate circuitry of decimation stage k+1.

Abstract: Active-on-active microelectronic devices are described. For example, a first die is on a second die with a bottom surface of a first substrate facing a top surface of a second substrate, respectively, to provide a die stack. The first and second dies each have metal layers in ILD layers to provide a first stack structure and a second stack structure, respectively. The first stack structure is interconnected to an upper end of a TSV of the first die. A metal layer of the second stack structure near a bottom surface of the first substrate is interconnected to a lower end of the TSV. A power distribution network layer of the second stack structure is located between lower and upper layers of the metal layers thereof. A transistor located at least in part in the second substrate is interconnected to the power distribution network layer to receive supply voltage or ground.

Abstract: Some examples described herein provide for a multi-chip structure including one or more memory dies stacked on a die having a programmable integrated circuit (IC). In an example, a multi-chip structure includes a package substrate, a first die, and a second die. The first die includes a programmable IC, and the programmable IC includes a memory controller. The first die is on and attached to the package substrate. The second die includes memory. The second die is stacked on the first die. The memory is communicatively coupled to the memory controller.

Abstract: Examples herein describe performing non-sequential DMA read and writes. Rather than storing data sequentially, a DMA engine can write data into memory using non-sequential memory addresses. A data processing engine (DPE) controller can submit a first job using first parameters that instruct the DMA engine to store data using a first non-sequential write pattern. The DPE controller can also submit a second job using second parameters that instruct the DMA engine to store data using a second, different non-sequential write pattern. In this manner, the DMA engine can switch to performing DMA writes using different non-sequential patterns. Similarly, the DMA engine can use non-sequential reads to retrieve data from memory. When performing a first DMA read, the DMA engine can retrieve data from memory using a first sequential pattern and then perform a second DMA read where data is retrieved from memory using a second non-sequential read pattern.

Abstract: Disclosed herein is an optical filter configured for wavelength division and multiplexing capable of transmitting and receiving signals. The optical filter includes an optical waveguide configured to receive at an input multiple signals with different wavelengths. The optical filter includes a plurality of channels coupled at different locations along a length of the optical waveguide. Each of the plurality of channels is configured to transmit a respective one of the multiple signals. A number of ring filter stages in a first channel of the plurality of channels that is closer to the input of the optical waveguide is greater than a second channel in the plurality of channels further away from the input of the optical waveguide.

Abstract: Approaches for simulating a circuit include receiving simulation input data from a testbench executing on a computer system by a simulator interface executing on the computer system. The simulator interface receives simulation output data the according to a hardware bus protocol specified by a simulated circuit for communication and simulates handshaking with the simulated circuit according to the hardware bus protocol in response to receiving the simulation input data and simulation output data. The simulator interface provides the simulation input data to the simulated circuit by according to the hardware bus protocol and provides the simulation output data to the testbench.

Abstract: A circuit for implementing an input/output connection in an integrated circuit device is described. The circuit comprises a pull-up circuit comprising a first plurality of transistors coupled in series, wherein a gate of a first transistor of the first plurality of transistors is configured to receive a first dynamic bias signal; a pull-down circuit comprising a second plurality of transistors coupled in series, the pull-down circuit being coupled to the pull-up circuit at an output node, wherein a gate of a first transistor of the second plurality of transistors is configured to receive a second dynamic bias signal; and an input/output contact coupled to the output node. A circuit for implementing an input/output connection in an integrated circuit device including a splitter circuit for receiving an input signal on an input pad is also described. A method of implementing an input/output connection in an integrated circuit device is also described.

Abstract: An example method of implementing a quantized neural network (QNN) for a programmable device includes: identifying multiply-accumulate operations of neurons in the QNN; converting the multiply-accumulate operations to memory lookup operations; and implementing the memory lookup operations using a pre-compute circuit for the programmable device, the pre-compute circuit storing a pre-computed output of a neuron in the QNN for each of the memory lookup operations.

Abstract: Embodiments herein describe adapting a PIM model to compensate for changing PIM interference. A PIM model can include circuitry that generates a PIM compensation value that compensates for (i.e., mitigates or subtracts) PIM interference caused by transmitting two or more transmitter (TX) carriers in the same path. The disclosed adaptive scheme generates updated coefficients for the PIM model which are calculated after the RX signal has been removed from the RX channel. In this manner, as the PIM interference changes due to environmental conditions (e.g., temperature at the base station), the adaptive scheme can update the PIM model to generate a PIM compensation value that cancels the PIM interference.

Abstract: Detection circuitry for an integrated circuit (IC) includes voltage divider circuitry, comparison circuitry, and calibration circuitry. The voltage divider circuitry receives a power supply signal and output a first reference voltage signal and a supply voltage signal based on the power supply signal. The comparison circuitry compares the first reference voltage signal and the supply voltage signal to generate an output signal. The calibration circuitry alters one or more parameters of the voltage divider circuitry to increase a voltage value of the supply voltage signal based on the comparison of the first reference voltage signal with the supply voltage signal.

Abstract: Embodiments herein describe a hardware accelerator (e.g., a network acceleration engine) for a blockchain machine or node. The hardware accelerator parses packets containing separate components of a block of transactions to generate data to perform a validation process. To avoid the latency that comes with using software, the embodiments herein describe a protocol processor in the hardware accelerator that parses the packets and prepares the data so it can be consumed by downstream components in the accelerator without software intervention. These downstream components can then perform a validation operation to validate one or more transactions before those transactions are committed (i.e., added) to a ledger of a permissioned blockchain.

Abstract: Synchronizing system resources of a multi-socket data processing system can include providing, from a primary System-on-Chip (SOC), a trigger event to a global synchronization circuit. The primary SOC is one of a plurality of SOCS and the trigger event is provided over a first sideband channel. In response to the trigger event, the global synchronization circuit is capable of broadcasting a synchronization event to the plurality of SOCS over a second sideband channel. In response to the synchronization event, the system resource of each SOC of the plurality of SOCS is programmed with a common value. The programming synchronizes the system resources of the plurality of SOCS.

Abstract: A system includes a bridge circuit configured for low latency communication among integrated circuits (ICs). The bridge circuit includes a plurality of transceiver circuits. Each transceiver circuit is coupled to a corresponding parallel channel in the IC. Each transceiver circuit is configured to send and receive data over the corresponding parallel channel. Each transceiver circuit includes a transmit channel configured to packetized data received from the corresponding parallel channel for transmission over a serial link to a second IC. Each transceiver circuit includes a receive channel configured to depacketize data received from the serial link from the second IC. The serial link is asynchronous to each of parallel channel coupled to the first bridge circuit.

Abstract: Processing of a neural network specification includes gathering first layers of a neural network graph into groups of layers based on profiled compute times of the layers and equalized compute times between the groups. Each group is a subgraph of one or more of the layers of the neural network. The neural network graph is compiled into instructions for pipelined execution of the neural network graph by compute circuits. The compiling includes designating, for each first subgraph of the subgraphs having output activations that are input activations of a second subgraph of the subgraphs, operations of the first subgraph to be performed by a first compute circuit and operations of the second subgraph to be performed by a second compute circuit. The compute circuits are configured to execute the instructions.

Abstract: One example of the present disclosure is an integrated circuit (IC). The IC includes an inverter with an input and an output, a clock transmission gate coupled to the output of the inverter; and a plurality of storage cells. The clock transmission gate is coupled to each of the plurality of storage cells, wherein each of the plurality of storage cells comprises a plurality of nodes arranged based on a minimum spacing.

Abstract: A system can include a microprocessor having a prefetch queue including a plurality of slots configured to store program counter values (PCVs) and instructions, a pipeline configured to receive instructions from the prefetch queue, and a select circuit coupled to the prefetch queue. The select circuit may selectively freeze a first slot of the plurality of slots and selectively output a frozen PCV and a frozen instruction from the first slot while frozen. The microprocessor can include write logic coupled to the prefetch queue and a comparator circuit coupled to the prefetch queue and the select circuit. The write logic may load data into unfrozen slots of the prefetch queue. The comparator circuit may compare a target PCV with the frozen PCV to determine a match. The select circuit indicates, to the pipeline, whether the frozen instruction is valid based on the comparing.

Abstract: Approaches for placing logic of a circuit design include determining respective relative activation rates of control paths in a high-level language (HLL) program by a design tool. The HLL program specifies a circuit design. The design tool compiles the HLL program into logic functions and determines respective relative activation rates of signal connections between the logic functions based on the relative activation rates of the control paths in the HLL program. The design tool selects placement locations on an integrated circuit device for the logic functions using a placement cost minimization function that factors the relative activation rates of the signal connections into placement costs.