where the voltage is regulated and lowered once more, to the required operating voltage of the load. In a FPA implementation, the sequence is reversed. Regulation occurs first, and the regulation
module can be placed in the optimal position for space utilization and thermal management. A regulated voltage approaching 48V is bussed across the circuit to the transformation module, which performs what we refer to as current multiplication,
adjacent to the load. Bussing high voltage minimizes the current levels across the circuit, thereby minimizing the potential for distribution losses and reducing the volume of the conduit (e.g., the copper wire). Placing the relatively low noise,
low heat current multiplication module adjacent to the load further minimizes the potential for distribution losses associated with bussing a low operating voltage to the load and reduces the potential influence of the power system on the
performance of the load.
A typical FPA implementation for delivering 48V DC from a server backplane to a 1.0V microprocessor would consist
of three modules: a PRM (Pre-Regulator Module) regulator, a VTM (Voltage Transformation
Module) current multiplier, and a proprietary communications controller. In contrast, a commodity Intermediate Bus Architecture (IBA) design for delivering 48V DC from a server backplane to a 1.0V microprocessor requires an additional
conversion stage, to reduce 48V to 12V, and, at the point of load, a voltage regulation module (i.e., a VRM consisting of multiple switching regulators, each representing a phase and consisting of two switching transistors, one or more
capacitors, and an inductor, with the transistors switched by pulse width modulation controller). For a 200W two stage, multiphase application, a 12V commodity IBA implementation would require an intermediate bus converter, to reduce 48V to 12V, and
a VRM solution consisting of parallel phases (i.e., multiple switching regulators) to reduce and regulate the current for use at 1.0V by the microprocessor. Such a commodity IBA implementation requires a significantly higher component count,
consumes more motherboard area, requires more copper conduit, generates more heat due to switching and distribution losses, offers inferior dynamic response, and can be meaningfully less efficient than a 48V FPA implementation.
The advantages of FPA over legacy power distribution architectures are most evident in high performance computing applications. Our
Power-on-Package power system solutions meet the computational performance requirements of artificial intelligence (AI). The microprocessors typically used in AI, particularly in more computationally demanding machine
learning or training applications, are graphics processing units (GPUs) and custom application-specific integrated circuits (ASICs). Unlike central processing units (CPUs), which are designed for
serial execution of complex and broad instruction sets, GPUs and AI ASICs are designed for massively parallel (i.e., concurrent) processing of repetitive transactions or calculations. As such, GPUs and AI ASICs generally operate at processing
frequencies requiring the higher levels of average and peak current delivered by our FPA-based solutions. Our most popular Power-on-Package solution, consists of one MCD© (Modular Current Driver) unit, providing high-bandwidth, low-noise regulation, and two MCM©
(Modular Current Multiplier) units, providing high performance current multiplication. Power-on-Package delivers unprecedented current levels to GPUs and AI ASICs, in part due to the placement of the MCMs directly on the substrate onto which the
processor is mounted, thereby minimizing distribution losses associated with high current levels. Placement of MCM units on the substrate also reduces the number of GPU or ASIC processor substrate pins required for power, allowing for their use by
other functions (e.g., memory input/output (I/O)). This three-module laterally-mounted Power-on-Package configuration, powering an AI accelerator card requiring 350W, delivers 0.7V, 650A average current, and up to 1,200A peak current to
the GPU or AI ASIC, with superior transient response and unmatched power density.
Our latest innovation for powering processors is
vertical power delivery, which involves mounting our highest-performance solutions on the underside of the motherboard, opposite the GPU or AI ASIC, thereby enabling a further reduction in distribution losses at the load, yielding higher efficiency
and unprecedented power density. Vertically-mounting the solution allows unrestricted access to microprocessor input/output I/O pins on the top side of the motherboard, thereby improving I/O speed and memory access, which are a priority for GPUs and
AI ASICs in AI applications. We have completed the final development stages of our vertical power delivery solutions and began shipping released products to customers in 2022.
Our proprietary technologies enable us to offer a range of Advanced Products, in various package formats across functional families,
applicable to other market segments and power distribution architectures other than FPA. Within computing, these market segments include AC to DC voltage conversion and DC voltage distribution in server racks and high voltage conversion across
datacenter infrastructure. We also offer Advanced Product power system solutions for aerospace and aviation (e.g., for use in satellites, unmanned aerial vehicles, and various airframes, including battery-powered aircraft, for which small size,
light weight, and design flexibility are advantageous); defense electronics (e.g., for use in airborne, seaborne, or field communications and radar, for which reliability in harsh environments is a priority); industrial automation, instrumentation,
and test equipment (e.g., for use in robotics and semiconductor testing, for which high power levels and precision performance are required); solid state lighting (e.g., for use in large scale displays and signage, for which, again, small size,
light weight, and design flexibility are advantageous); telecommunications and networking infrastructure (e.g., for use in high-throughput data distribution and pole-mounted small-cell base stations); and vehicles (e.g., in autonomous driving
applications, electric vehicles, and hybrid electric vehicles).