Abstract: A method and apparatus for adding and multiplying floating-point operands such that a fixed-size mantissa result is produced. In accordance with the present addition method, the mantissa of a first floating-point operand is shifted in accordance with relative operand exponent information. Next, the first operand mantissa is added to the second operand mantissa. The addition step includes replacing a least significant non-overlapped portion of the first operand mantissa with a randomly-generated carry-in bit. In accordance with the multiplication method, a partial product array is generated from a pair of floating-point operand mantissas. Next, prior to compressing the partial product array into a compressed mantissa result, a lower-order bit portion of the partial product array is replaced with a randomly generated carry-in value.

Abstract: A cascaded test circuit with inter-bitline drive devices for evaluating memory cell performance provides for circuit delay and performance measurements in an actual memory circuit environment. A row in a memory array is enabled along with a set of drive devices that couple each bitline pair to the next in complement fashion to form a cascade of memory cells. The drive devices can be inverters and the inverters can be sized to simulate the bitline read pre-charge device and the write state-forcing device so that the cascade operates under the same loading/drive conditions as the operational with memory cell read/write circuits. The last and first bitline in the row can be cascaded, providing a ring oscillator or the delay of the cascade can be measured in response to a transition introduced at the head of the cascade. Weak read and/or weak write conditions can be measured by selective loading.

Abstract: A digital circuit with dynamic power and performance control via per-block selectable operating voltage level permits dynamic tailoring of operating power to processing demand and/or compensation for process variation. A set of processing blocks having a power supply selectable from two different power supply voltage levels is provided. The power level of the overall circuit is set by selecting the power supply voltage for each block to yield a combination of blocks that meets operating requirements. Alternatively, one circuit per pair from a set of pairs of redundant logic blocks supplied by the different power supply voltage levels can be selected to meet the operating requirements. The unselected blocks can be disabled by disabling foot devices or disabling transitions at the inputs to the unselected blocks. Performance measurement and feedback circuits can be included to tune the power consumption and performance level of the circuit to meet an expected level.

Abstract: Methods and apparatus are provided for varying one or more of a supply voltage and reference voltage in an integrated circuit, using independent control of a diode voltage in an asymmetrical double-gate device. An integrated circuit is provided that is controlled by one or more of a supply voltage and a reference voltage. The integrated circuit comprises an independently controlled asymmetrical double-gate device to adjust one or more of the supply voltage and the reference voltage. The independent control may comprise, for example, a back gate bias. The independently controlled asymmetrical double-gate device may be employed in a number of applications, including voltage islands, static RAM, and to improve the power and performance of a processing unit.

Abstract: A single-stage level shifting circuit is used to interface control signals across the boundary between voltage domains with differing positive or ground voltage potentials Asserted states are determined by the difference between the positive voltages potentials and the ground potentials. A lower positive power supply potential is not used to turn OFF PFET coupled to a higher positive power supply potential. Likewise a higher ground power supply potential is not used to turn OF NFETs coupled to a power domain where is significant ground shift. The single stage level shifting circuit has keeper devices that hold asserted states using voltages within the power gated domain.

Abstract: A dynamic logic gate has an asymmetrical dual-gate PFET device for charging a dynamic node during a pre-charge phase of a clock. A logic tree evaluates the dynamic node during an evaluate phase of the clock. The front gate of the asymmetrical dual-gate PFET device is coupled to the clock signal and the back gate is coupled to the ground potential of the power supply. When the clock is a logic zero both the front gate and the back gate are biased ON and the dynamic node charges with maximum current. The clock signal transitions to a logic one during the evaluation phase of the clock turning OFF the front gate. The back gate remains ON and the asymmetrical dual-gate PFET device operates as a keeper device with a current level sufficient to counter leakage on the dynamic node.

Abstract: A ring oscillator is formed using inverting stages configured from asymmetrical dual gated FET (ADG-FET) devices. The simplest form uses an odd number of CMOS inverter stages configured with an ADG-PFET and an ADG-NFET. The front gates are used as the logic inputs and are coupled to preceeding outputs from the main ring. The back gates of the ADG-PFET devices are coupled to a first control voltage and the back gates of the ADG-NFET devices are coupled to a second control voltage that is the complement of the first control voltage referenced to an off-set voltage. Other configurations of logic inverting stages using ADG-FET devices may also be used. The control voltage is varied to modulate the current level set by the logic state at the inputs coupled to the front gates.

Abstract: A dynamic logic gate has a device for charging a dynamic node during a pre-charge phase of a clock. A logic tree evaluates the dynamic node with a device during an evaluate phase of the clock. The dynamic node has a keeper circuit comprising an inverter with its input coupled to the dynamic node and its output coupled to the back gate of a dual gate PFET device. The source of the dual gate PFET is coupled to the power supply and its drain is coupled to the dynamic node forming a half latch. The front gate of the dual gate PFET is coupled to a logic circuit with a mode input and a logic input coupled back to a node sensing the state of the dynamic node. The mode input may be a slow mode to preserve dynamic node state or the clock delayed that turns ON the strong keeper after evaluation.

Abstract: A dynamic logic gate has a dynamic node pre-charged in response to a pre-charge phase of a clock signal and a logic tree with a plurality of logic inputs for evaluating the dynamic node during an evaluate phase of the clock signal in response to a Boolean combination of the logic inputs. The logic tree has a stacked configuration with at least one multi-gate FET device for coupling an intermediate node of the logic tree to the dynamic node in response to a first logic input of the plurality of logic inputs or in response to the pre-charge phase of the clock signal. The multi-gate FET device has one gate coupled to the first logic input and a second gate coupled to a complement of the clock signal used to pre-charge the dynamic node.

Abstract: An LSDL circuit replaces the normal clock control of the pre-charge device for the dynamic node with a control signal that is logic zero whenever the circuit is in an active mode and is a logic one when the circuit is in standby mode. The pre-charge device holds the dynamic node at a pre-charged logic one state independent of the clock. During the logic one evaluate time of the clock, the logic tree determines the asserted state of the dynamic node. During the evaluate time, the asserted state is latched by the static LSDL section. The dynamic node then re-charges to the pre-charge state. Since the pre-charge device is not de-gated during the evaluate time, the dynamic node cannot be inadvertently discharged by noise causing an error. Likewise, since the clock does not couple to the pre-charge device a load is removed from the clock tree lowering clock power.

Abstract: A digital transmission circuit and method providing selectable power consumption via multiple weighted driver slices improves the flexibility of an interface while reducing transmitter power consumption, area and complexity when possible. A cascaded series of driver stages is provided by a set of parallel slices and a control logic that activates one or more of the slices, which combine to produce a cascaded active driver circuit. The power consumption/drive level selectability of the slice combination provides a driver that can be fine-tuned to particular applications to provide the required performance at a minimum power consumption level.

Abstract: A digital transmission circuit and method providing selectable power consumption via single-ended or differential operation improves the flexibility of an interface while reducing power consumption when possible. A differential path is provided through the transmitter output driver stages and portions are selectively disabled when the transmission circuit is in a lower-power operating mode. A single-ended to differential converter circuit can be used to construct a differential signal for output to the final driver stage. The selection of power mode can be made via feedback from a channel quality measurement unit or may be hardwired or selected under programmatic control. The longer delay or skew of the lower-power single-ended mode is compensated for by the relaxed requirements of the channel when conditions permit the use of the lower-power single-ended mode.

Abstract: A multi-threshold complementary metal-oxide semiconductor (MTCMO) bus circuit reduces bus power consumption via a reduced circuit leakage standby and pulsed control of standby mode so that the advantages of MTCMOS repeater design are realized in dynamic operation. A pulse generator pulses the high-threshold voltage power supply rail standby switching devices in response to changes detected at the bus circuit inputs. The delay penalty associated with leaving the standby mode is overcome by reducing cross-talk induced delay via a cross-talk noise minimization encoding and decoding scheme. A subgroup of bus wires is encoded and decoded, simplifying the encoding, decoding and change detection logic and results in the bus subgroup being taken out of standby mode only when changes occur in one or more of the subgroup inputs, further reducing the power consumption of the overall bus circuit.

Abstract: A buffer, logic circuit, and data processing system employing fast turn-off drive circuitry for reducing leakage. Leakage current in logic circuitry is managed by coupling and decoupling the voltage potentials applied to large, high-leakage devices. Circuitry includes a low leakage logic path for holding logic states of an output after turning off high-leakage devices. A fast turn-off logic path in parallel with the low leakage logic path is used to assert each logic state in the forward direction from input to output. The large output device in each fast turn-off path is relieved of leakage stress by asserting logic states at driver inputs that cause the driver to turn OFF after the output logic state has been asserted.

Abstract: A low power consumption pipeline circuit architecture has power partitioned pipeline stages. The first pipeline stage is non-power-gated for fast response in processing input data after receipt of a valid data signal. A power-gated second pipeline stage has two power-gated modes. Normally the power rail in the power-gated second pipeline stage is charged to a first voltage potential of a pipeline power supply. In the first power gated mode, the power rail is charged to a threshold voltage below the first voltage potential to reduce leakage. In the second power gated mode. the power rail is decoupled from the first voltage potential. A power-gated third pipeline stage has its power rail either coupled to the first voltage potential or power-gated where its power rail is decoupled from the first voltage potential. The power rail of the second power-gated pipeline stage charges to the first voltage potential before the third power-gated pipeline stage.

Abstract: Virtual power-gated cells (VPC) are configured with control circuitry for buffering control signals and a power-gated block (PGB) comprising two or more NFETs for virtual ground rail nodes and PFETs for virtual positive rail nodes. Each VPC has a control voltage input, a control voltage output, a node coupled to a power supply voltage potential, and a virtual power-gated node that is coupled and decoupled from the power supply potential in response to logic states on the control input. The control signals are buffered by non-power-gated inverters before being applied to the input of a PGB. VPCs may propagate a control signal that is in phase with or inverted from a corresponding control signal at the control input. VPCs may be cascaded to create virtual power rails in chains and power grids. The control signals are latched at the cell boundaries or latched in response to a clock signal.

Abstract: Leakage current in logic circuitry is managed by coupling and decoupling the voltage potentials of the power supply from circuitry with large high leakage devices. Driver circuits comprise a low leakage logic path for holding logic states of the output. A high leakage logic path in parallel with the low leakage logic path is used to assert each logic state in the forward direction from input to output. The large output device in each high leakage path that enhances the current drive of a logic state on the output are leakage stress relieved by allowing their drive inputs to collapse after the output logic state has been asserted. The high leakage logic paths employ multiple stages with collapsing logic states that are generated in response to asserted logic states on the output and logic states of the low leakage logic path thus reducing the device sizes needed to control leakage.

Abstract: A buffer/driver having large output devices for driving multiple loads is configured with three parallel paths. The first logic path is made of small devices and is configured to provide the logic function of the buffer without the ability to drive large loads. Second and third logic paths have the logic function of the first logic path up to the last inverting stage. The last inverting stage in each path is a single device for driving the logic states of the buffer output. The second and third logic paths have power-gating that allows the input to the pull-up and pull-down devices to float removing gate-leakage voltage stress. When the second and third logic paths are power-gated, the first logic path provides a keeper function to hold the logic state of the buffer output. The buffer may be an inverter, non-inverter, or provide a multiple input logic function.

Abstract: A method and system for interactive modeling of high-level network performance with low-level link design provides a tool for optimizing networked computing systems and their link components simultaneously. The method models a fixed portion of a network and specifies operational performance levels and power constraints. A solution is chosen for a non-fixed network portion and the network is simulated to determine link requirements and synthesizes links in conformity with the link requirements. The links are analyzed to determine performance (e.g., bandwidth) and requirements (e.g., power) and network performance is recalculated. An iterative loop from the selection of the non-fixed topology through synthesis and recalculation of link performance can be implemented to optimize the link and network design.

Abstract: A buffer/driver having large output devices for driving multiple loads is configured with three parallel paths. The first logic path is made of small devices and is configured to provide the logic function of the buffer/driver without the ability to drive large loads. Second and third logic paths have the logic function of the first logic path up to the last inverting stage. The last inverting stage in each path is a single device for driving the logic states of the buffer output. The second and third logic paths have power-gating that allows the input to the pull-up and pull-down devices to float removing gate-leakage voltage stress. When the second and third logic paths are power-gated, the first logic path provides a keeper function to hold the logic state of the buffer output. The buffer/driver may be an inverter, non-inverter, or provide a multiple input logic function.