Design Article
PCM Progress Report No 4: Brains
Ron Neale
7/22/2011 2:32 PM EDT
This phase-change memory (PCM) progress report explores the use of a PCM to emulate a component of the brain, the synapse, in an impressive piece of work from Stanford University.
While many have claimed and suggested that amorphous memory and threshold switches might be able to emulate functions in the neural network, this latest work from Stanford is a serious device-based physical example [1]. The experimental evidence presented supports the claim by the authors that, to their knowledge, “this is the first demonstration of a single element electronic synapse with the capability of the modulation of the time constant and the realization of the different STDP (Spike-timing Dependant Plasticity) kernels.” If, going forward, the dreams of neural network emulation are to be fully realized, the challenges to PCM device designers in terms of precision, discrimination and scaling will exceed, by far, anything that has been accomplished to date.
Linking PCM and synapse
As illustrated in Figure 1, the synapse is a complex connector that is involved in controlling the passage of neural messages from one neuron to the next. It is part of the learning process, it functions by combing and remembering the timing of events (spikes) in the two neurons that it links, called the pre-synaptic and post-synaptic neurons.
Also shown in Figure 1 is the PCM device structure that was chosen for this experimental work. Readers familiar with the trials and tribulations of the attempts at commercializing PCM will recognize this as the familiar “mushroom,” “dome,” or “barrel” structure, with or without heater electrode. The PCM emulation devices used have a contact diameter of 75nm and from the bottom up use a W-TiN-xtal GST- TiN structure.
In the brain, the synapse learning process is facilitated by STDP of the synapse. While the focus here will be on the PCM performance and achievements, it might be useful to relate the terms used in the study of brain activity with PCM electronics. These are shown in Table 1.
The synapse is an important component because, in the human brain, it is estimated there are some 1015 synapses connecting some 1011 neurons in a three-dimensional network.
The form of the synapse characteristic that must be emulated is shown in Figure 2. The vertical axis is the learned weight of the connection while the horizontal axis relates to the time that a post neuron activity (spike) precedes or lags a pre-neuron activity (spike), as shown in the insets to Figure 2, resulting in the curves to the left or right of the vertical axis respectively.
While many have claimed and suggested that amorphous memory and threshold switches might be able to emulate functions in the neural network, this latest work from Stanford is a serious device-based physical example [1]. The experimental evidence presented supports the claim by the authors that, to their knowledge, “this is the first demonstration of a single element electronic synapse with the capability of the modulation of the time constant and the realization of the different STDP (Spike-timing Dependant Plasticity) kernels.” If, going forward, the dreams of neural network emulation are to be fully realized, the challenges to PCM device designers in terms of precision, discrimination and scaling will exceed, by far, anything that has been accomplished to date.
Linking PCM and synapse
As illustrated in Figure 1, the synapse is a complex connector that is involved in controlling the passage of neural messages from one neuron to the next. It is part of the learning process, it functions by combing and remembering the timing of events (spikes) in the two neurons that it links, called the pre-synaptic and post-synaptic neurons.
Also shown in Figure 1 is the PCM device structure that was chosen for this experimental work. Readers familiar with the trials and tribulations of the attempts at commercializing PCM will recognize this as the familiar “mushroom,” “dome,” or “barrel” structure, with or without heater electrode. The PCM emulation devices used have a contact diameter of 75nm and from the bottom up use a W-TiN-xtal GST- TiN structure.
In the brain, the synapse learning process is facilitated by STDP of the synapse. While the focus here will be on the PCM performance and achievements, it might be useful to relate the terms used in the study of brain activity with PCM electronics. These are shown in Table 1.
The synapse is an important component because, in the human brain, it is estimated there are some 1015 synapses connecting some 1011 neurons in a three-dimensional network.
The form of the synapse characteristic that must be emulated is shown in Figure 2. The vertical axis is the learned weight of the connection while the horizontal axis relates to the time that a post neuron activity (spike) precedes or lags a pre-neuron activity (spike), as shown in the insets to Figure 2, resulting in the curves to the left or right of the vertical axis respectively.
|
|
|
|
|
Navigate to related information
janine.love
7/22/2011 2:53 PM EDT
I asked Ron Neale to take a look at some of the current work being done in PCM and he took the time to analyze some work being done by Stanford University. Look for more on PCM, with a new progress report coming in a week or two.
Sign in to Reply
iniewski
8/2/2011 6:35 PM EDT
Could you let us know where the report is published/posted? Kris
Sign in to Reply
resistion
7/22/2011 9:10 PM EDT
I think the desired outcome will be foiled by resistance drift. I also wonder if something similar could be accomplished by Flash? If not, the reason(s) would also be important.
Sign in to Reply
R G.Neale
7/23/2011 6:38 AM EDT
Resistion-You could be right, I guess it's up to someone to prove the point experimentally, arm waving solutions and claims of brain functions or any other aspect of PCM developments are no longer acceptable.
The 100 level resolution (1%)for Flash might result in some drift. In a modern scaled Flash how many electrons does that involve for each step? It is even possible if all devices in an emulated synapse network drift together the relative learned experience of the neural network is the same, others will have to answer that question. Drift tolerance is the feature that IBM use in their MLC-PCM. Both IBM-MLC and PCM-Onyx were originally part of this Progress Report I understand that material will now appear in the next PCM Progress Report, #5 in the near future.
Sign in to Reply
resistion
7/24/2011 12:02 AM EDT
The drift can be compensated with a reference, but each new data input needs a new reference, so things (costs, power, delays) add up that way.
100 levels in flash I think is not doable, even 8 levels or 3 bits is borderline. Too few random electrons to tell the difference.
Sign in to Reply
janine.love
7/23/2011 4:44 PM EDT
I have hidden some of the comments on this article because of their unproductive nature. My thanks to those readers who pointed them out. I understand and am glad that we are passionate about our work, but let's keep it professional. Thanks.
Sign in to Reply
Volatile Memory
7/25/2011 10:58 AM EDT
Dear RF/Memory Editor:
Yes, let's keep it professional! When pseudo-research is touted as some kind of breakthrough in a respected publication, the duty of the editor is to notice, not to silence the whistleblower. The fact is, Mr. Neale dropped the ball on this one. He knew or should have known that Mr. Ovshinsky has claimed similar "results" for at least 25 years. Those claims and results turned out to be fraudulent. As will the latest "results" from the "researchers" at Stanford University.
Sign in to Reply
janine.love
7/25/2011 1:18 PM EDT
Debate and conversation is welcome. In fact, it is the intent of the comments section. As long as you refrain from name calling, I'm happy to let the comments stand and the debate ensue.
Sign in to Reply
Solster
7/25/2011 5:55 PM EDT
The fact remains that this is an academic paper peer-reviewed by credentialed research scientists and a professor. Constructive dialogue is the cornerstone of academic research, while "commentators" to web-articles expressing opinions with no apparent personal credibility whatsoever, don't contribute much to the debate, really. There's hopefully a reason why R.G. Neale (and not a certain "commentator") was invited to write this review article and any real constructive debate on this paper could really just be a rebuttal paper in Nano Letters. Anything less, especially those without any technical discussions and instead full of dubious accusations, is worth little more than idle chit-chat for mere entertainment.
Sign in to Reply
Volatile Memory
7/25/2011 9:05 PM EDT
Solster: Care to explain how exactly Mr. Ovshinsky managed to publish a fraudulent paper, describing the 16-level magic neuron device, on the pages of the peer-reviewed Japanese Journal of Applied Physics in 2004:
http://jjap.jsap.jp/link?JJAP/43/4695/
I wonder what Mr. Neale has to say about it.
Sign in to Reply
rbtbob
7/25/2011 12:44 AM EDT
Some of the readers that are not current on the research being done on phase change materials and devices in the last few years might like to read some of the papers presented at the European Phase Change and Ovonics Science Symposium. Regarding the subject of Mr. Neale's analysis, I recommend the paper presented by Stan Ovshinsky at the 2004 Symposium.
http://www.epcos.org/library/library2010.htm
Sign in to Reply
Volatile Memory
7/25/2011 11:10 AM EDT
Here is the 10x Microsofts quote, accompanied with the "results:"
http://goo.gl/WCxQm
The document was created in December of 2004 (and published in early 2005) when Mr. Ovshinsky was still at the helm of Ovonic Cognitive Computer and its parent.
Sign in to Reply
R G.Neale
7/25/2011 5:23 AM EDT
rbtbob-I was careful to put the precedence claim in quotes in case I had missed a paper. My brief was to explore what the Stanford team had been able to get the PCM to do based on their real experimental data, not to research the whole field of bio-science for claims and counter claims. The word "promising" in the title of the paper you recommend gives cause for concern. I think to date the whole field of phase change memory has been beset and damaged by too many unfulfilled promises.
Sign in to Reply
R G.Neale
7/26/2011 10:57 AM EDT
Solster: Outside of the subject of this paper, and your comment""The fact remains that this is an academic paper peer-reviewed by credentialed research scientists and a professor. Constructive dialogue is the cornerstone of academic research, while "commentators" to web-articles expressing opinions with no apparent personal credibility whatsoever, don't contribute much to the debate, really.""
I think as the web takes over paper publishing peer group review will change and become web based with an opportunity for peer group review to come quickly from all quarters, including your professors etc. With the editors responsible for the removal of any offensive material. I am convinced that is the future of peer group review.
Also would you care to explain your words regarding myself and my reputation in parenthesis and quotes.
Sign in to Reply
R G.Neale
7/26/2011 12:01 PM EDT
Solster- On the subject of peer group review there is an interesting article "Putting Peer Review on Trial", by Raoul Franklin,in PhysicsWorld, December 2010,p17.Published by the Institute of Physics (IoP)
Key quote "The system we use to judge our peers work must be made more transparent" He points out a number of problems and defects with the present method and offers some possible changes.
If you read it it may shake your faith a little in the present system.I will try and assess if it is on the web.
Sign in to Reply