How are the T-1000's programming and memory stored?

Question

Watching Terminator 2 again, I find the statement

"The T-1000 can't form complex machines"

to be intriguing, mainly in the sense that the T-1000 itself is a complex machine. This got me thinking about how the T-1000 operates as a machine, in particular the computing side of it, and specifically how data is stored within it. It is clear that the T-1000 contains some very sophisticated programming (by our standards). Its abilities include:

• participating in natural language conversations (to at least some extent)
• mimicing voices that it hears
• reshaping its appearance to nearly perfectly match humans that it only briefly encounters
• seeking and acquiring information required to track a target, and strategizing for the disposal of that target
• operating a wide variety of machinery: cars, trucks, helicopters, etc.

The first ability above is a significant marker of the programming sophistication behind the T-1000, but so too is the third: it implies that its programming includes extremely robust algorithms for convincingly simulating the mechanics of a person's gait, visage, skin, and clothing from only brief samplings of these components' motion.

Question:

How are the T-1000's programming (almost certainly very extensive) and its memory stored within the liquid metal? Without moving parts, how are particular routines or memory files accessed? Related to this, when the T-1000 is separated into many little blobs, how are its programming and memory recombined? If data is distributed diffusely throughout the liquid metal, would the T-1000 suffer data loss (or display programming errors) if one of the blobs were lost?

This would give a nice explanation as to why he waits for the last blob to rejoin his foot before resuming the attack on John Connor and the T-800 near to the end of T2.

In-universe information is preferred, including facts gleaned from novelizations, DVD extras, interviews, etc.

Answer 1

According to the Randall Frakes novelisation, the T-1000's memory (and its programming and its 'molecular DNA') are stored across its entire structure:

It didn’t have a wafer-circuit brain to think with. It was something on a completely new level of artificial intelligence. The molecular brain acted like the rest of the thing, a liquid. And now it bubbled with possibilities.

All of them lethal.

The distant sound of approaching sirens reached its auditory sensors, which could have been formed anywhere on its body (since every molecule had the “genetic” blueprints for all needed parts programmed into them), but were now in the shape of human ears.

as to what the main programming of the separated parts was;

In the corridor, the T-1000’s head was lying in two mutilated masses on its shoulders. The concept of pain had never factored into the sensory sphere of the liquid machine. Pain was an indicator of damage to a part of the organism.

But this “organism” didn’t have parts, except on the molecular level. And its molecules were each primitive, miniaturized versions of the total machine. If any section were parted, the separated halves would revert to metal poly-alloy. The only default command it had in molecular memory was to find the main mass again and rejoin it. Each molecule had a range of fourteen kilometers. And the blasted apart sections of the T-1000 were much closer than that. So, after a moment of hesitation caused by ballistic shock, it rapidly reformed into a healthy human face, with absolutely no trace of “injury.”