Not So Fast

Dropping the voltage may make lots of sense in a planar chip, but how about in a stacked die?

popularity

Semiconductor engineering has always been about taking two steps forward and one step back. Or, if you’re more cynical, you don’t really move at all. You just get better at what you do.

The cynical approach is that nothing is really new because it’s all been done before. In effect, 2.5D packages are multi-chip modules with a better business case and more advanced versions of what existed a couple decades ago. Virtualization was created in the 1960s. Software written more tightly to the hardware is a repeat of the RISC vs. CISC battle. The list goes on.

But stacking of die in a 2.5D stack and ultimately into 3D stacks offers some new wrinkles that have never been explored. While there has been lots of talk about dropping the voltage in chips to improve battery life in mobile devices, or energy efficiency in those systems with a plug, it’s not that simple.

Most large chipmakers believe the realistic limit for voltage reduction in most devices is somewhere between 0.8 and 0.6 volts, at least over the next decade. There are a few that will push it further, at least in theory. It’s difficult enough to hit these numbers in existing planar chips because the leakage goes up as the voltage goes down. Some devices, including transistors, may have to be redesigned to use this lower voltage, or improve on efficiency in other ways such as going vertical with FinFETs. But with some great engineering and lots of really smart people, this is achievable—at least in theory.

It gets even harder, though, in a stacked die where all the pieces aren’t built by one company—and the whole attraction of stacked die is that you don’t have to build it all yourself. You have to integrate the pieces. And when they leak out of the middle or the sides of the chip because the voltage is too low and the other pieces aren’t equipped to handle that leakage it can cause big problems. In fact, dropping the voltage can kill a multi-layer chip, if it’s not thought through carefully.

In this case, it may be wiser to look for power reduction in other areas first and approach a voltage drop much more cautiously. When it comes to stacking and voltage, it’s better to take multiple small steps forward than one giant step in the wrong direction.