On the advent of the memristor
The first question you might ask is, "what's a memristor?" Basically, it's a solid state device that alters its resistance based on the most recent voltage applied to it. In its fundamental form, it can be used as a digital switch. That functionality is a one device, drop-in replacement for about 10 transistors in a CMOS latch (non-volatile digital memory), making it smaller, lighter and cheaper. You can read more detail about HP's development of the memristor here (or for more scientific review, here).
While it's possible that the first implementation of memristor-based memory will be in a CMOS chip, it seems like a daunting task. The device HP developed requires an Aluminum oxide substrate (doped appropriately). Unfortunately, most CMOS devices don't work well with an aluminum base, and it's also somewhat resistant to x-ray lithography (how they put circuits onto the silicon). Parceling the substrate into different materials is (I'm told) nigh-impossible.
It occurred to me that there is an immediate solution: epitaxial deposition. This method of chip production has been relegated to analog signal processing (ASP), because it generally produces larger devices (which can handle more current). Epitaxial is also best suited to JFET ("GaAs circuit") production; JFETs in turn, are well-suited to ASP - due to extremely fast response times.
It has been shown however, that JFETs can be used in a NOR/NOR configuration (rather than NAND/NAND like CMOS transistors) to produce identical digital results as their CMOS counterparts. The two main drawbacks (if I recall correctly) are: the lack of a low-power latch and a higher cost of production. Alternate substrate material is not exceptional in Epitaxial production, and memristors are a zero-power latch.
The response times of JFETs make them speed-comparable to CMOS circuits at 1/10th of their scale. Certain digital circuit 'tricks' would have to be re-invented for use with a JFET processor, like using two back-to-back memristors for a reversible diode. I'm confident that a memristor flash drive could be put to market inside 2 years - not the 5+ years HP has slated for development of multi-substrate CMOS lithography.
This course of action also leads directly into HP's desire for a "synapse-like" circuit - that would effectively produce a programmable analog computer.
The development of a memristor is going to be a huge boon to computers eventually. But, given HP's predilection for a dogmatic approach to marketing, I'm not sure we should hold our breath waiting for it to come.
(Ok, ok... so this post is 7 months after the fact. Sue me.)
While it's possible that the first implementation of memristor-based memory will be in a CMOS chip, it seems like a daunting task. The device HP developed requires an Aluminum oxide substrate (doped appropriately). Unfortunately, most CMOS devices don't work well with an aluminum base, and it's also somewhat resistant to x-ray lithography (how they put circuits onto the silicon). Parceling the substrate into different materials is (I'm told) nigh-impossible.
It occurred to me that there is an immediate solution: epitaxial deposition. This method of chip production has been relegated to analog signal processing (ASP), because it generally produces larger devices (which can handle more current). Epitaxial is also best suited to JFET ("GaAs circuit") production; JFETs in turn, are well-suited to ASP - due to extremely fast response times.
It has been shown however, that JFETs can be used in a NOR/NOR configuration (rather than NAND/NAND like CMOS transistors) to produce identical digital results as their CMOS counterparts. The two main drawbacks (if I recall correctly) are: the lack of a low-power latch and a higher cost of production. Alternate substrate material is not exceptional in Epitaxial production, and memristors are a zero-power latch.
The response times of JFETs make them speed-comparable to CMOS circuits at 1/10th of their scale. Certain digital circuit 'tricks' would have to be re-invented for use with a JFET processor, like using two back-to-back memristors for a reversible diode. I'm confident that a memristor flash drive could be put to market inside 2 years - not the 5+ years HP has slated for development of multi-substrate CMOS lithography.
This course of action also leads directly into HP's desire for a "synapse-like" circuit - that would effectively produce a programmable analog computer.
The development of a memristor is going to be a huge boon to computers eventually. But, given HP's predilection for a dogmatic approach to marketing, I'm not sure we should hold our breath waiting for it to come.
(Ok, ok... so this post is 7 months after the fact. Sue me.)