RF heterogenous ICs are making headway

Article By : Jean-Jaques (JJ) DeLisle

This article looks at the current state of the RF heterogeneous integrated circuit (HIC) industry and the outlook on the technology in the next few years.

In the last RF blog, we discussed heterogeneous integrated circuit (HIC) technology and how that pertained to developments with RF semiconductor devices. The previous blog mainly consisted of background information and some of the trends driving RF HICs. This blog will focus more on the current state of the industry and the outlook on RF HIC technology in the next few years. Several industry experts were interviewed regarding this topic, and there have been recent product releases of HICs with RF technology that demonstrate that RF HICs are now a reality.

To recap, HICs are a synthesis of several ICs (chiplets) either within the same package on the same plane (2D), stacked within a package via an interposer (2.5), or stacked directly on top of each other (3D). There are other technologies being developed, such as direct transistor stacking, but that is outside the scope of this discussion. There are also technologies being developed that would enable class III-V semiconductor and CMOS combinations, which are also outside this discussion’s scope. The focus here is mainly on 2.5D and system-in-package (SiP) 2D technologies, as that ‘s what is currently available and in the works.

The main drivers for this development are attempts to expand the capability of advanced/active antenna systems (AAS) for telecommunication applications like 5G, satellite, and military/aerospace communications as well as radar and electronic warfare (EW) applications. The reason deeper levels of integration are desirable is that the relative size, weight, and power (SWaP) of more tightly integrated chiplets and IC technologies has the potential to be substantially superior.

Then, there are performance bottlenecks associated with traditional 2D technologies such as PCB and low/high temperature co-fired ceramic (LTCC/HTCC). Moreover, connectorized packaging suffers from even greater interconnect losses and space efficiency compared to IC technology. In order to increase throughput, add additional antenna elements and processors, or deploy more advanced technologies on the edge, the current 2D and connectorized options are clearly facing limitations.

The first RF HIC?

A recently launched case is the AMD-Xilinx Versal AI Edge adaptive compute acceleration platform (ACAP). It’s composed of chiplets for multiple methods of processing/computing, RF data converters, power converters, memory, and communication buses. As such, this is essentially a complete programmable network-on-a-chip. The various chiplets that comprise this RF SiP are produced by different vendors and likely required substantial development efforts to integrate all of the available technologies over a several year collaborative development process.

A Versal AI Edge device has enabled a cacheless-memory hierarchy to ease memory access bottlenecks. Source: AMD-Xilinx

In this case, the ACAP is likely 2D, possibly 2.5D in some cases, but still presents a much smaller footprint than a PCB version of a device with the same features. The description of the capability of this platform sounds quite impressive as a do-all solution.

However, this platform, though highly programmable, is inflexible from a hardware perspective compared to PCB or connectorized technologies. The DSP/FPGA and RF converters—both analog-to-digital converters (ADCs) and digital-to-analog-converters (DACs)—are fixed within the hardware and cannot be modified or upgraded. In fact, it may be virtually impossible to rework or repair such units outside of a specialized manufacturing facility. Furthermore, the development tools for this platform are extremely specialized, so it may not be possible to port the development system to another platform if that’s ever desired.

Upsides and downsides

This example hits at the crux of deeper levels of integration. Though enhancing SWaP, efficiency, and other metrics is possible with enough investment in developing RF HIC technology, it comes at the cost of flexibility. Here is a comment from Christopher Marki, CEO of Marki Microwave:

“The main drawback is SiPs assume some amount of specificity in application/use. A generic single function block like a broadband amp can be used in many places. Putting that amp into a mixer SiP reduces the application space of that amp to being used as a converter. So, integration always sacrifices the number of customers who can use a certain solution. But, that new solution may be significantly more appealing to certain customers, and unlock more potential upsides. SiP does not solve all problems, there is a trade that balances technical and business considerations.”

Marki Microwave recently announced a line of mixers with integrated LO-drivers: broadband LO driver amplifiers. As discussed earlier, though less complex than the AMD-Xilinx ACAP, even integrating two components that are frequency paired in RF circuits limits the flexibility of the components in the SiP. Hence, there needs to be an application with enough demand to fuel the development of such technology.

In discussion with other RFIC/MMIC vendors, it seems clear that there is a general move toward developing RF SiPs in the near term, with potential for more advanced RF HIC integration in the future. There is still a divide between the digital/processing/conversion RF HIC electronics and RF front-end (RFFE) technology.

There may soon be developments that include more of the RFFE in the form of chiplets, most likely encompassing class III-V semiconductors. The European Space Agency (ESA) has issued a fund for such development, specifically targeted toward Ka-band AAS applications. There are also a host of similar DARPA projects relating to including RFFE technology in RF HICs.


This article was originally published on Planet Analog.

Jean-Jaques (JJ) DeLisle, an electrical engineering graduate (MS) from Rochester Institute of Technology, has a diverse background in analog and RF R&D, as well as technical writing/editing for design engineering publications. He writes about analog and RF for Planet Analog.


Leave a comment