Stackable chips pave the way for reconfigurable hardware

Recognising the challenges posed by the permanent nature of semiconductors, MIT researchers have demonstrated a new interconnect technology that allows for chips to be stacked as needed at any point in time. Why is non-configurable hardware a problem, what did the team present, and how does it demonstrate the power of optical connections?

Why is non-configurable hardware a problem?

As technology improves, so does the demand for new technologies, and this demand helps to fund the development of ever-better technologies. While this cycle has helped rapidly expand humanity’s capabilities over the last century, it has also caused numerous challenges, including dwindling resources, dependency on technology (and thus potential political instability), and massive environmental damage.

With regards to environmental damage, the rapid expansion of technological development can see devices made redundant after just a few years of use and, as such, can often require consumers to frequently replace electronic devices. This rapid replacement of devices results in large amounts of e-waste whose size seems to only increase with time. As e-waste contains a wide variety of harmful compounds, incorrectly disposing of electronic devices can result in contaminated soil and groundwater and a serious impact on human health.

One solution to this problem would be configurable hardware that would allow users to only replace key components of a design instead of having to get an entirely new device. For example, a reconfigurable smartphone could have the processor replaced with a newer version, RAM could be increased in size with the addition of a second RAM module, and internal storage could also be upgraded. As the core components of a smartphone don’t require upgrading (i.e., screen, enclosure, and buttons), e-waste could be significantly reduced. Furthermore, removing old parts increases the chance of recycling, and these older components could even find their way into other devices (i.e., IoT upgrades).

But trying to achieve this with current technology is extremely challenging for a multitude of reasons. With regards to integrated circuits on PCBs, most are soldered into place, meaning that they cannot be easily removed. Desktop PCBs use sockets, but sockets in a mobile device are impractical. Secondly, integrated circuits cannot be upgraded with ease as they require metal interconnects, which are impossible to sever. Thirdly, even if a semiconductor could have its top layer removed in favour of a new chip stack, it would require extreme alignment of metal layers which is something that would be impossible outside of a controlled lab.

FPGAs and CPLDs do present a viable option for reconfigurable hardware, but their large cost and size make them difficult to implement for mobile devices. Even though they are reprogrammable, they are unable to have programmable logic cell densities sufficient to create advanced circuits whose performance is on par with the latest semiconductor device. An FPGA can be programmed as an ARM CPU, for example, but this CPU would be larger and more power-hungry than a dedicated ARM processor in silicon.

MIT researchers develop stackable chips built like LEGO®

Recently, researchers from MIT have developed a new stackable chip technology that allows for a semiconductor device to be upgraded over time without needing to remove the chip in favour of a new one.

Initially, the researchers set the goal of creating an image recognition system that would combine an image sensor, an interconnecting layer, and a network of artificial synapses. This three-layer block is then programmed to recognise one of three characters (M, I, and T), and three of these blocks are then layered on top of each other (to allow for the detection of all three characters).

These layers are connected to each other through optical link layers consisting of an LED and associated photosensor. Using an optical link removes the need for electrical wires and allows layers to be removed and added as needed (similar to LEGO®).

The advantage of this setup was recognised when their AI chip struggled to correctly identify I and T characters. Instead of creating an entirely new chip, the researchers instead swapped out the processor layer in favour of a denoising layer that improved the AI’s ability to recognise I and T characters.

How does this use of optical layers demonstrate the power of stackable chips?

The ability to replace processing layers of a semiconductor to improve its capabilities represents a serious development that could change how hardware is constructed in the future. Optical connections are not only easier to connect and disconnect (as there is no need for a physical connection), but they can also handle significantly more bandwidth and operate at greater frequencies than standard electrical connections. As such, future devices that utilise optical interconnects instead of electrical will not only be faster but also be reconfigurable with ease.

But what the MIT researchers have demonstrated also bodes well for creating environmentally-friendly devices (or at least help reduce e-waste). Instead of replacing devices outright, semiconductor layers could be stacked onto existing chips using optical interconnects to add new functionality. For example, a user who specifically needs graphical capabilities could integrate an additional GPU core, while a user who wants improved security could add a security layer.

Overall, the ability to stack semiconductor layers with added functionality using optical connections presents the engineering world with all kinds of possibilities. If these components are combined with optical PCBs, then we could see a dramatic shift from electrical connections to optical in the near future.