12/07/2026
Tillie Sutton overcomes the Von Neumann bottleneck.
A post about Tillie’s new chip.
As conscientious readers of my blog you will be familiar with Tillie Sutton, but like me, you may not have known about the Von Neumann bottleneck. After all, ‘performance lag in traditional computer architecture due to the separation of the processor (CPU) and computing memory’ is not always forefront in our minds.
The Von Neumann bottleneck happens to be really important. In computer systems data and instructions travel back and forth along a shared communication pathway called the bus. That process slows the system’s overall speed, depending how fast this data can be transferred. The CPU and the memory are two distinct hardware units. Whilst instructions and data share the bus, computers can typically only do one thing at at a time – either fetch an instruction, or read/write data. As processors calculate results much faster than the bus can feed them data, this forces the CPU to sit idle while waiting for information to arrive. It is this roadblock that limits computers – until now.
____________________________________________________________
Last week Alistair and I received messages from the Sutton Syndicate indicating that they had some important news.
‘We are sending Tillie back to school,’ wrote J, ‘it’s a retrofit which will impress,’ he continued, somewhat muddying our minds by mixing his metaphors.
‘I think I might know what this is,’ said Alistair with a knowing look. ‘I bet its a cerebellum-inspired electronic chip.’
‘Until recently the Von Neumann bottleneck created a major roadblock in high-performance computing,’ Alistair explained. ‘As the complexity of software has grown, especially to meet the needs of artificial intelligence, this constant back-and-forth movement of huge datasets has created not only severe processing delays but involved high energy consumption.’
‘Some engineers at Northwestern University in Illinois have discovered a way of getting round the problem – in fact they have pinched it from humans and it uses 10,000 times fewer computer operations than current AI requires.’ ‘They have already tried it out on testing abnormal heart rhythms with 98% accuracy.’
Half-understanding I propose, ‘let’s TEAMS with the Sutton Syndicate to test your theory.’ ‘Perhaps they will even allow Tillie to explain it herself?’
__________________________________________________
Tillie Sutton appears promptly on our screen. Today she is not robed, but dressed casually and seated in her chambers library.
‘Your human brain has two main parts,’ she explains, ‘the cerebrum at the top which controls higher cognitive functions, voluntary movement, conscious thought, reasoning, memory, and sensory interpretation – and the cerebellum at the bottom which handles more instinctive responses such as motor coordination, posture, balance, and motor learning.’
‘While most neuromorphic computing tries to recreate the thought-heavy cerebrum, last week I was fitted with a device that emulates the human reflex-driven cerebellum. It ignores expected data and reserves my energy for the unexpected. It’s a bit like the human gatekeeper you, J and Alistair have – an organic novelty detector, filtering out routine background information from conscious activity, and firing up only when something unexpected comes up.’
‘What it means is that I no longer need to shuttle data around all the time. With my new memtransistors I can use my memory and logic operations in the same physical space at the same time, slicing my energy demands.’
Nodding to J, Alistair asks Tillie, ‘How did the guys at Northwestern manage to do this?
‘They developed atomically thin molybdenum disulfide film, and partially overlapping the semiconductor with one electrode, they reversed the voltage direction to toggle the device between unexpected and expected states,’ she replies. ‘In other words, routine input doesn’t trigger active processing, just as when we watch a sunset we don’t think about Rayleigh Scattering.’
J intervenes, ‘in clinical trials the memtransistor ignored thousands of routine heartbeats, whilst the moment an arrhythmia occurred, it flagged the defect within milliseconds, processing more than twice as fast as conventional AI models. For Tillie, this means, like us, she can focus on the novel – rather than process the normal, cutting down the need to connect to energy-hungry cloud data centres.’
Unexpectedly, Alistair turned back to Tillie to ask her a further question.
‘How does it feel to have the new chip,’ he inquires.
‘It makes life and reasoning so much easier now I don’t have to process all of the background noise,’ she responds. ‘It’s just like leaving work and getting on the bus to go home,’ she adds with a smile. https://stephentwist.substack.com/p/tillie-sutton-overcomes-the-von-neumann
